CN117593301B

CN117593301B - Machine vision-based memory bank damage rapid detection method and system

Info

Publication number: CN117593301B
Application number: CN202410073849.5A
Authority: CN
Inventors: 尹春
Original assignee: Shenzhen Oscoo Tech Co ltd
Current assignee: Shenzhen Oscoo Tech Co ltd
Priority date: 2024-01-18
Filing date: 2024-01-18
Publication date: 2024-04-30
Anticipated expiration: 2044-01-18
Also published as: CN117593301A

Abstract

The application relates to the technical field of machine vision, and discloses a memory bank damage rapid detection method and system based on machine vision. The method comprises the following steps: performing image detection on the target memory bank to obtain a first memory area characteristic image; performing storage area feature detection weight calculation to obtain a second storage area feature image; detecting the damage of the storage area through an initial memory bank damage detection model to obtain a storage area damage detection frame; performing feature decomposition of the damage assembly to obtain S first damage feature region graphs and K second damage feature region graphs, performing feature enhancement and image fusion on the damage regions, and generating a target damage region enhanced image; marking the information of the damaged component and outputting the result to obtain a memory bank damage detection result; network parameter updating and optimizing are carried out to obtain a target memory bank damage detection model, and the method improves the accuracy of rapid memory bank damage detection.

Description

Machine vision-based memory bank damage rapid detection method and system

Technical Field

The application relates to the technical field of machine vision, in particular to a memory bank damage rapid detection method and system based on machine vision.

Background

With the continuous development of computer technology and the expansion of application fields, the memory bank is one of key components of computer hardware, and the performance and reliability requirements are also higher and higher. However, since the memory bank is affected by physical damage or external environment during use, an effective detection method is required to identify the potential problem of the memory bank in time. The quick memory bank damage detection method based on machine vision is a popular field of research and industrial application due to the non-contact, high efficiency and automation characteristics. The method can realize rapid detection and analysis of damage while guaranteeing the performance of the memory bank by utilizing the computer vision and deep learning technology, and is expected to improve the reliability and stability of a computer system.

There are still some challenges and problems with the prior art. The variety of memory bank damage types, including physical damage, circuit failure, etc., is still a complex problem how to effectively identify and classify different types of damage. Secondly, the detection of the memory bank damage needs to consider the influence of environmental factors, such as illumination, angles and the like, which cause instability of the detection result. In the existing research, how to realize the optimization and optimization of the network parameters of the memory bank damage detection model so as to improve the performance and the robustness of the memory bank damage detection model also needs more intensive research and exploration. Therefore, further research and improvement are still needed for a quick memory bank damage detection method based on machine vision so as to meet the increasing detection demands of computer hardware.

Disclosure of Invention

The application provides a method and a system for quickly detecting memory bank damage based on machine vision, which are used for improving the accuracy of quickly detecting the memory bank damage.

In a first aspect, the present application provides a method for quickly detecting memory bank damage based on machine vision, where the method for quickly detecting memory bank damage based on machine vision includes:

performing image detection on a target memory bank to obtain an initial memory bank image, and performing storage region feature extraction on the initial memory bank image to obtain a first storage region feature image;

Performing detection environment analysis on the initial memory bank image to obtain detection environment parameter data, and performing storage area feature detection weight calculation on the first storage area feature image according to the detection environment parameter data to obtain a second storage area feature image;

Inputting the second storage area characteristic image into a preset initial memory bank damage detection model, and detecting storage area damage through a storage area damage detection network in the initial memory bank damage detection model to obtain a storage area damage detection frame;

Performing damage component feature decomposition on the second storage region feature image according to the storage region damage detection frame through a damage feature extraction network in the initial memory bank damage detection model to obtain S first damage feature region images and K second damage feature region images, and performing damage region feature enhancement and image fusion on the S first damage feature region images and the K second damage feature region images to generate a target damage region enhanced image, wherein S and K are positive integers, and S=3K;

Performing damage component information labeling and result output on the target damage region enhanced image through a damage labeling network in the initial memory bank damage detection model to obtain a memory bank damage detection result;

And performing network parameter optimization on the initial memory bank damage detection model according to the memory bank damage detection result by adopting a particle swarm optimization algorithm to obtain a network parameter optimization set, and performing network parameter updating and optimization on the initial memory bank damage detection model according to the network parameter optimization set to obtain a target memory bank damage detection model.

In a second aspect, the present application provides a machine vision-based memory bank damage rapid detection system, where the machine vision-based memory bank damage rapid detection system includes:

the detection module is used for carrying out image detection on the target memory bank to obtain an initial memory bank image, and carrying out storage region feature extraction on the initial memory bank image to obtain a first storage region feature image;

The computing module is used for carrying out detection environment analysis on the initial memory bank image to obtain detection environment parameter data, and carrying out storage area feature detection weight computation on the first storage area feature image according to the detection environment parameter data to obtain a second storage area feature image;

The input module is used for inputting the second storage area characteristic image into a preset initial memory bank damage detection model, and carrying out storage area damage detection through a storage area damage detection network in the initial memory bank damage detection model to obtain a storage area damage detection frame;

The decomposition module is used for carrying out damage component feature decomposition on the second storage region feature image according to the storage region damage detection frame through a damage feature extraction network in the initial memory bank damage detection model to obtain S first damage feature region images and K second damage feature region images, carrying out damage region feature enhancement and image fusion on the S first damage feature region images and the K second damage feature region images, and generating a target damage region enhanced image, wherein S and K are positive integers, and S=3K;

The marking module is used for marking the damage component information and outputting the result of the target damage region enhanced image through a damage marking network in the initial memory bank damage detection model to obtain a memory bank damage detection result;

and the optimization module is used for performing network parameter optimization on the initial memory bank damage detection model according to the memory bank damage detection result by adopting a particle swarm optimization algorithm to obtain a network parameter optimization set, and performing network parameter updating and optimization on the initial memory bank damage detection model according to the network parameter optimization set to obtain a target memory bank damage detection model.

In the technical scheme provided by the application, the image detection and the storage area characteristic extraction are carried out on the memory strip, so that the focus of damage detection is placed on a key area. The localization method greatly improves the detection efficiency and reduces the waste of computing resources, thereby completing the memory bank damage detection more rapidly. By detecting environment analysis and storing area characteristic detection weight calculation, the method can adapt to different detection environments and conditions. The memory bank damage detection method has the advantages that the memory bank damage detection is more robust, the high efficiency can be maintained under various illumination, angles and environments, and false alarms and missing alarms are reduced. And adopting the feature decomposition and the image fusion of the damage assembly to realize the multi-level analysis of the damage of the memory bank. The method is helpful for more accurately identifying and positioning different types of damage, so that the detection result is more detailed, and more targeted information is provided for subsequent maintenance or replacement. By adopting the particle swarm optimization algorithm, the network parameters of the memory bank damage detection model can be automatically optimized and optimized to adapt to different damage types and environmental conditions, so that the performance and generalization capability of the detection model are improved. Through the damage labeling network, not only can the damage of the memory bank be detected, but also the damage assembly can be labeled and analyzed in detail, so that the accuracy of the rapid detection of the damage of the memory bank is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained based on these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a diagram illustrating an embodiment of a method for quickly detecting memory bank damage based on machine vision according to an embodiment of the present application;

Fig. 2 is a schematic diagram of an embodiment of a memory bank damage rapid detection system based on machine vision according to an embodiment of the present application.

Detailed Description

The embodiment of the application provides a memory bank damage rapid detection method and system based on machine vision. The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments described herein may be implemented in other sequences than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus.

For easy understanding, the following describes a specific flow of an embodiment of the present application, referring to fig. 1, and an embodiment of a method for quickly detecting memory bank damage based on machine vision in the embodiment of the present application includes:

Step S101, performing image detection on a target memory bank to obtain an initial memory bank image, and performing storage region feature extraction on the initial memory bank image to obtain a first storage region feature image;

It can be understood that the execution body of the present application may be a machine vision-based memory bank damage rapid detection system, and may also be a terminal or a server, which is not limited herein. The embodiment of the application is described by taking a server as an execution main body as an example.

Specifically, image detection is performed on the target memory bank. And acquiring the image data of the memory bank through a proper image acquisition device or a camera. This initial memory stripe image contains various interference factors, such as noise. And preprocessing the initial memory stripe image to improve the accuracy of subsequent analysis. A noise removal operation is performed to eliminate interference in the image. This helps to clearly capture the characteristics of the memory bank. Edge enhancement is performed to highlight the edge structure of the memory stripe, making it easier to analyze. After image enhancement, luminance parameter correction is performed. Luminance correction parameter data is calculated to ensure that the luminance levels of the images are uniform. At the same time, color space conversion is performed to convert the color representation of the image into a form more suitable for analysis, and color conversion parameter data is acquired. Subsequently, a perceptual weighting assignment is made, which helps to better weigh the individual features in subsequent analysis. The calculation of the perceptual weighting parameter data determines the importance of the different features in the final feature parameter data set. And calculating to obtain a characteristic parameter data set of the memory bank according to the brightness correction parameter data, the color conversion parameter data and the perception weight parameter data. This data set will contain various features such as brightness, color, edges, etc. to describe the image characteristics of the memory stripe. And carrying out storage area parameter configuration and storage area feature extraction on the second memory bank image by utilizing the feature parameter data set. And configuring proper storage area parameters according to the content of the characteristic parameter data set so as to ensure that the storage area of the memory bank is fully analyzed. And performing feature extraction to acquire a first storage area feature image.

Step S102, carrying out detection environment analysis on an initial memory bank image to obtain detection environment parameter data, and carrying out storage area feature detection weight calculation on a first storage area feature image according to the detection environment parameter data to obtain a second storage area feature image;

Specifically, detection environment analysis is performed on the initial memory bank image. Including obtaining initial environmental parameter data and a loss detection environmental impact set. These environmental parameters include light conditions, temperature, humidity, etc., while the loss detection environment influence factor set is used to record various factors that influence damage detection, such as illumination changes, temperature fluctuations, etc. And according to the loss detection environment influence factor set, carrying out environment parameter analysis on the initial environment parameter data, and understanding how the environment factors influence the quality and characteristics of the memory bank image. For example, if the ambient light is weak, the image processing algorithm needs to be adjusted to accommodate low light conditions. And further processing the detected environmental parameter data by adopting a preset analytic hierarchy process. This includes building a hierarchy, typically including a target layer, a criteria layer, and a factor layer. In this hierarchy, the target layer is the problem to be solved, i.e. how to adjust the storage area feature detection weights according to the environmental parameter data. The criteria layer includes criteria for evaluating different factors, while the factor layer includes various influencing factors such as illumination, temperature, etc. According to the hierarchical structure, a discrimination matrix is constructed, which is used for comparing the relative importance of the factors. This matrix allows for pairwise comparisons of different factors and derives weights based on the comparison results. These weights will reflect the extent to which each factor affects the detection of lesions. And obtaining a weight vector by carrying out mean value operation on a plurality of column vectors in the discrimination matrix. This vector contains weight information for each factor to indicate their relative importance in the overall environmental parameter. The weight vectors are normalized to ensure that their sum is equal to 1, resulting in storage area feature detection weight data. These weight data will be used to perform storage area feature detection weight calculation on the first storage area feature image, ultimately generating a second storage area feature image.

Step S103, inputting a second storage area characteristic image into a preset initial memory bank damage detection model, and carrying out storage area damage detection through a storage area damage detection network in the initial memory bank damage detection model to obtain a storage area damage detection frame;

Specifically, the second storage area characteristic image is input into a storage area damage detection network in the initial memory bank damage detection model. This network typically includes several key layers including an image normalization layer, a storage area damage anomaly detection layer, and an anchor frame layer. In the image normalization layer, the second storage area characteristic image is subjected to image size normalization processing, so that the input image is ensured to be matched with the requirement of a training model in size, and subsequent processing is performed. And detecting the storage area damage to the normalized storage area characteristic image through a two-layer convolution network in the storage area damage abnormal detection layer. The goal of these convolutional networks is to extract features in the image and identify the damage condition of the storage area. Through a plurality of convolution layers and activation functions, the network can learn representations of different damage features and generate probability values that each pixel belongs to a damage abnormal region. And generating a corresponding storage area damage abnormal probability map according to the probability value of each pixel belonging to the damage abnormal area. This probability map can be used to represent the extent of damage to different areas in the image. And judging the damage abnormal region of the storage region damage abnormal probability map according to a preset probability threshold. In general, a region where the pixel probability value is higher than the threshold value will be regarded as a damage abnormality region. And generating a corresponding storage area damage detection frame according to the positions and the shapes of the abnormal areas. The generated storage area damage detection frame can be used for identifying a damaged part on the memory strip, and a damage detection result is provided.

Step S104, performing damage component feature decomposition on the second storage region feature image according to the storage region damage detection frame through a damage feature extraction network in the initial memory bank damage detection model to obtain S first damage feature region images and K second damage feature region images, and performing damage region feature enhancement and image fusion on the S first damage feature region images and the K second damage feature region images to generate a target damage region enhanced image, wherein S and K are positive integers, and S=3K;

Specifically, the second storage area characteristic image is input into a damage characteristic extraction network in the initial memory bank damage detection model. This network is typically composed of several key layers, including a damage component feature decomposition layer, an improved laplacian, and an image fusion layer. At the lesion assembly feature decomposition level, the second storage area feature image is decomposed into S first lesion feature area maps and K second lesion feature area maps. This decomposition is accomplished by the convolution and pooling layers of the network, separating out the different damaged components or feature areas in the image. In the improved Laplace operator layer, S first damage characteristic region graphs and K second damage characteristic region graphs are processed to capture detailed information of the damage characteristic regions. This step helps to extract the tiny features of the lesion area, thereby more accurately identifying and describing the lesion. And generating a characteristic region map fusion rule according to the storage region damage detection frame. This rule is used to determine how to perform lesion field feature enhancement on the S first lesion field maps and the K second lesion field maps. This may be weighted according to the relative importance of the different lesion characterization to generate a more accurate lesion field characterization. In the image fusion layer, the S first damage characteristic region graphs and the K second damage characteristic region graphs which are subjected to characteristic enhancement processing are subjected to image fusion, a plurality of target characteristic enhancement images are generated, and each image contains damage information with different degrees. These target feature enhanced images are generated to better demonstrate various aspects and degrees of memory bank damage, thereby helping the system to more accurately determine the type and severity of the damage.

Step S105, marking the damage component information and outputting the result of the target damage region enhanced image through a damage marking network in the initial memory bank damage detection model to obtain a memory bank damage detection result;

Specifically, the damage component information labeling is performed on the target damage region enhanced image through a damage labeling network in the initial memory bank damage detection model. The lesion marking network is typically a deep learning model that identifies and marks different lesion components in the image, such as cracks, scratches, defects, etc. Through training of the network, the shape, location and severity of the different damaged components can be understood and reflected in the target damaged component mask. And carrying out local damage optimization on the enhanced image of the target damage region according to the target damage component mask. And further processing each damaged part in the image according to the labeling information of the damaged component so as to improve the visibility and the identification of the damage. Including contrast enhancement, edge enhancement, noise removal, etc. And outputting a result of the target local optimization image subjected to the damage local optimization processing to generate a memory bank damage detection result. This result will include information about the location, shape, and severity of each damaged component. Based on this information, the system can accurately understand the damage on the memory bank and take appropriate action, such as repair or replacement of the damaged portion.

And S106, performing network parameter optimization on the initial memory bank damage detection model according to a memory bank damage detection result by adopting a particle swarm optimization algorithm to obtain a network parameter optimization set, and performing network parameter updating and optimization on the initial memory bank damage detection model according to the network parameter optimization set to obtain a target memory bank damage detection model.

Specifically, the result feedback is performed on the memory bank damage detection result based on a preset intelligent feedback mechanism, so as to obtain target feedback information. This feedback information typically includes assessment and improvement suggestions for performance of the initial memory bank damage detection model. By continuously collecting and analyzing the damage detection results, key information about the performance of the model can be obtained. And carrying out model parameter range analysis on the initial memory bank damage detection model according to the target feedback information so as to obtain a model parameter range set. The method is helpful for determining the reasonable range of the model parameters, avoiding too large parameter search space and improving the efficiency of the particle swarm optimization algorithm. And generating random initial values of a plurality of model parameters of the initial memory bank damage detection model through the model parameter range set to obtain a corresponding random initial value set. This set contains different initial value combinations of multiple model parameters for constructing particle swarms. And constructing a particle population for the random initial value set by using a preset particle population optimization algorithm. A particle group is a population of individuals (particles), each particle representing a set of values of a model parameter. The particles move in the search space and model parameters are continuously adjusted according to the adaptability of the particles so as to find the optimal solution. And calculating the particle fitness of the particle population to obtain a particle fitness set corresponding to the particle population. The fitness is typically calculated based on performance metrics of the lesion detection results, such as accuracy, recall, etc. The particle swarm algorithm determines an optimal solution by comparing fitness of different particles. Subsequently, iterative calculations and optimization solutions are performed on the set of particle fitness to find the best model parameter combination. The iterative process of the particle swarm algorithm continuously updates the model parameters until a stopping condition (such as the maximum number of iterations or performance convergence requirement) is met. And according to the network parameter optimization set, updating and optimizing network parameters of the initial memory bank damage detection model to obtain a target memory bank damage detection model. The optimization process can significantly improve the performance of the model, so that the model can more accurately detect the damage on the memory bank and provide more reliable detection results.

In the embodiment of the application, the image detection and the storage area characteristic extraction are carried out on the memory strip, so that the focus of damage detection is placed on the key area. The localization method greatly improves the detection efficiency and reduces the waste of computing resources, thereby completing the memory bank damage detection more rapidly. By detecting environment analysis and storing area characteristic detection weight calculation, the method can adapt to different detection environments and conditions. The memory bank damage detection method has the advantages that the memory bank damage detection is more robust, the high efficiency can be maintained under various illumination, angles and environments, and false alarms and missing alarms are reduced. And adopting the feature decomposition and the image fusion of the damage assembly to realize the multi-level analysis of the damage of the memory bank. The method is helpful for more accurately identifying and positioning different types of damage, so that the detection result is more detailed, and more targeted information is provided for subsequent maintenance or replacement. By adopting the particle swarm optimization algorithm, the network parameters of the memory bank damage detection model can be automatically optimized and optimized to adapt to different damage types and environmental conditions, so that the performance and generalization capability of the detection model are improved. Through the damage labeling network, not only can the damage of the memory bank be detected, but also the damage assembly can be labeled and analyzed in detail, so that the accuracy of the rapid detection of the damage of the memory bank is improved.

In a specific embodiment, the process of executing step S101 may specifically include the following steps:

(1) Performing image detection on the target memory bank to obtain an initial memory bank image, performing noise removal on the initial memory bank image to obtain a first memory bank image, and performing edge enhancement on the first memory bank image to obtain a second memory bank image;

(2) Performing brightness parameter correction on the second memory bank image to obtain brightness correction parameter data, performing color space conversion on the second memory bank image to obtain color conversion parameter data, and performing perception weight distribution on the second memory bank image to obtain perception weight parameter data;

(3) Obtaining a memory bank characteristic parameter data set according to the brightness correction parameter data, the color conversion parameter data and the perception weight parameter data;

(4) And carrying out storage area parameter configuration and storage area feature extraction on the second memory bank image according to the memory bank feature parameter data set to obtain a first storage area feature image.

Specifically, image detection is performed on the target memory bank to obtain an initial memory bank image. A clear image of the memory bank is captured by a high resolution camera and appropriate lighting conditions. The captured image contains various noise, such as noise introduced by factors such as uneven illumination, imperfect camera sensors, and the like. To improve the accuracy of subsequent processing, these noises need to be removed, which is typically achieved by using digital image processing techniques such as median filtering, gaussian filtering, etc. After noise is removed, the obtained first memory bank image is clearer and is suitable for further analysis. And then, carrying out edge enhancement processing on the first memory bank image. Edge enhancement is a technique that improves the visual effect and the feature saliency of an image by highlighting edge information in the image, and common methods include the use of sobel filters, laplace operators, and the like. The image after the edge enhancement (the second memory bank image) is more beneficial to identifying and analyzing potential damages such as micro cracks, scratches and the like of the memory banks. And correcting the brightness parameter of the second memory bank image. Brightness correction is the adjustment of the brightness of an image in order to maintain visual consistency of the image under different viewing conditions. This may be achieved by adjusting the histogram of the image or using a specific brightness correction algorithm. By this correction, it is possible to ensure that the brightness of the image matches the visual appearance of the actual memory bank. At the same time, color space conversion is performed to convert the image from one color space (e.g., RGB) to another (e.g., HSV or YCbCr) to better analyze and process the image data. The color space conversion is helpful to highlight the specific color characteristics of the memory bank, so that the damage detection is more accurate. And then, performing perceptual weight distribution on the second memory bank image. Different parts in the image are assigned different weights according to their importance in lesion detection. For example, the connection points or microcircuits of the memory banks are more vulnerable than other parts, and therefore higher weights are assigned to these areas. Through the steps, the memory bank characteristic parameter data set can be obtained, which is a group of data sets containing brightness correction parameters, color conversion parameters and perception weight parameters, and provides basis for subsequent storage area parameter configuration and characteristic extraction. And carrying out storage area parameter configuration and storage area feature extraction on the second memory bank image according to the feature parameter data sets, so as to obtain a first storage area feature image. In this process, machine learning algorithms, such as convolutional neural networks, can be used to automatically identify and extract key features of the memory bank, such as solder joints, chip locations, etc.

In a specific embodiment, the process of executing step S102 may specifically include the following steps:

(1) Detecting environment analysis is carried out on the initial memory bank image to obtain initial environment parameter data, and a loss detection environment influence factor set is obtained;

(2) Performing environmental parameter analysis on the initial environmental parameter data according to the loss detection environmental impact factor set to obtain detection environmental parameter data;

(3) Carrying out environmental impact factor hierarchical analysis on the detected environmental parameter data by a preset hierarchical analysis method, and constructing a hierarchical structure, wherein the hierarchical structure comprises a target layer, a criterion layer and a factor layer;

(4) Constructing a discrimination matrix for detecting environmental parameter data according to the hierarchical structure, and carrying out mean value operation on a plurality of column vectors in the discrimination matrix to obtain a weight vector;

(5) And carrying out normalization processing on the weight vector to obtain storage region feature detection weight data, and carrying out storage region feature detection weight calculation on the first storage region feature image according to the storage region feature detection weight data to obtain a second storage region feature image.

Specifically, detection environment analysis is performed on the initial memory bank image, and initial environment parameter data are obtained. Such data include lighting conditions, camera parameters (e.g., exposure time, ISO sensitivity), background noise levels, etc. Meanwhile, environmental factors affecting results in the damage detection process, such as temperature, humidity, vibration and the like, are collected, and collectively referred to as a damage detection environmental impact factor set. The initial environmental parameter data is analyzed according to the impact factor sets to obtain more accurate detection environmental parameter data. For example, if the detection is performed in a low light environment, the brightness and contrast of the image need to be considered; if vibration exists in the environment, image stability and the like need to be considered. This step requires the use of data analysis techniques, such as statistical analysis or machine learning methods, in order to extract valuable information from the raw data. And carrying out environmental impact factor hierarchical analysis on the detection environment parameter data through a preset hierarchical analysis method (ANALYTIC HIERARCHY Process, AHP), and constructing a hierarchical structure comprising a target layer, a criterion layer and a factor layer. This hierarchy is used to systematically and quantify the importance of various influencing factors in the detection environment. For example, the goal of the target layer is to optimize the accuracy of the lesion detection; the criterion layer focuses on factors such as illumination, stability, temperature and the like; at the factor level, specific parameters under each criterion, such as illumination intensity, vibration frequency, etc., are specifically analyzed. And constructing a discrimination matrix for detecting the environmental parameter data. This discriminant matrix is used to compare the relative importance between different factors, typically by data analysis. And obtaining a weight vector representing the relative importance of each environmental influence factor by carrying out average value operation on a plurality of column vectors in the discrimination matrix. This weight vector can guide the server which environmental parameters should be more focused on in the subsequent processing. And carrying out normalization processing on the weight vector to obtain storage area feature detection weight data. The normalization process ensures that the sum of the weight vectors is 1, which can be directly applied to the weighting of the influencing factors. And according to the storage area feature detection weight data, the server calculates the storage area feature detection weight of the first storage area feature image, so as to obtain a second storage area feature image. For example, if the lighting conditions are considered to be the most important factors affecting detection, more emphasis is placed on adjusting the brightness and contrast parameters of the image when feature extraction is performed. Also, if temperature is a critical factor, the color balance of the image needs to be adjusted to compensate for color deviation due to temperature variation.

In a specific embodiment, the process of executing step S103 may specifically include the following steps:

(1) Inputting the second storage area characteristic image into a storage area damage detection network in a preset initial memory stripe damage detection model, wherein the storage area damage detection network comprises an image standardization layer, a storage area damage abnormality detection layer and an anchor frame layer;

(2) Performing image size standardization processing on the second storage area characteristic image through an image standardization layer to obtain a standard storage area characteristic image;

(3) Performing storage area damage detection on the standard storage area characteristic image through a two-layer convolution network in the storage area damage abnormal detection layer to obtain a probability value that each pixel belongs to a damage abnormal area;

(4) And generating a corresponding storage area damage abnormal probability map according to the probability value that each pixel belongs to the damage abnormal area, and judging the damage abnormal area and generating a detection frame of the storage area damage abnormal probability map according to a preset probability threshold value to obtain the storage area damage detection frame.

Specifically, the characteristic image of the second storage area is input into a preset initial memory bank damage detection model. The storage area damage detection network in this model comprises three main components: an image standardization layer, a storage area damage abnormality detection layer and an anchor frame layer. The design and sequence of operations of this network provides a structural and automated framework for the overall lesion detection process, thereby ensuring efficient and accurate detection results. And processing the second storage area characteristic image through an image normalization layer. Image normalization is a preprocessing step that converts an input image into a format and size that can be efficiently processed by the model. This typically involves adjusting the size and scale of the images to ensure that they meet the requirements of the model input. Normalizing the processed image, i.e., the normalized storage area feature image, helps the model better understand and process image data from different sources and conditions. And analyzing the standard storage area characteristic image through the storage area damage abnormality detection layer. This layer contains two layers of convolutional networks, which are structures commonly used in deep learning, suitable for processing image data. The primary function of these two-layer convolutional networks is to identify specific features in the image, such as edges, textures, and shapes, which facilitate lesion detection. Through the processing of the two-layer convolution network, the probability value that each pixel belongs to the damage abnormal region can be obtained. For example, if a pixel is located on a crack or scratch in a memory bank, the probability that it belongs to an abnormal region of damage will be significantly higher than pixels in other normal regions. And generating a corresponding storage area damage abnormal probability map according to the probability value of each pixel belonging to the damage abnormal area. This probability map provides an important visual representation of subsequent lesion detection and localization, with the color or intensity of each pixel representing the likelihood that it is identified as a lesion abnormality. Further, the damage abnormal probability map is analyzed according to a preset probability threshold value, so that damage abnormal areas can be judged and identified. The probability threshold is set in relation to the sensitivity and specificity of the lesion detection, too high a threshold results in missed detection, and too low a threshold results in false detection. Based on the result of the determination of the damage abnormal region, a storage region damage detection frame is further generated. The detection frames are used for accurately positioning the areas with the damage on the memory strips, and provide a basis for subsequent damage characteristic extraction and analysis. For example, if a high probability of damaging anomaly region is detected at a corner of a memory bank, a detection box would be generated around this region, highlighting the region for further analysis.

In a specific embodiment, the process of executing step S104 may specifically include the following steps:

(1) Inputting the second storage area characteristic image into a damage characteristic extraction network in an initial memory bank damage detection model, wherein the damage characteristic extraction network comprises a damage component characteristic decomposition layer, an improved Laplace operator and an image fusion layer;

(2) Decomposing the second storage area characteristic image into S first damage characteristic area diagrams and K second damage characteristic area diagrams through a damage component characteristic decomposition layer;

(3) Capturing damage characteristic region details of the S first damage characteristic region graphs and the K second damage characteristic region graphs through an improved Laplacian operator;

(4) Generating a characteristic region map fusion rule according to the storage region damage detection frame, and carrying out damage region feature enhancement on S first damage characteristic region maps and K second damage characteristic region maps according to the characteristic region map fusion rule to obtain a plurality of target feature enhancement images, wherein S and K are positive integers, and S=3K;

(5) And carrying out image fusion on the plurality of target feature enhanced images through the image fusion layer to generate a target damage region enhanced image.

Specifically, the second storage area characteristic image is input into a damage characteristic extraction network in the initial memory bank damage detection model. The damage feature extraction network comprises a damage component feature decomposition layer, an improved Laplacian and an image fusion layer. The second storage area feature image is decomposed into a plurality of different lesion feature area maps by a lesion component feature decomposition layer. In this process, the lesion characterization layer identifies and separates out different lesion features in the image, such as cracks, scratches, or corroded areas. The working principle of this layer is to decompose the image into S first injury feature region maps and K second injury feature region maps by analyzing local features of the image, where S and K are predefined positive integers and S is three times K. For example, if a memory bank having micro-cracks and corrosion sites is being detected, the first damage signature would be centered on the cracks and the second damage signature would be centered on the corrosion sites. These lesion feature area maps are subjected to more detailed feature capture by the modified laplace operator. The laplace operator is an edge detection operator commonly used for image processing and is capable of highlighting the high frequency features of an image, i.e. the rapidly changing parts of the image. The purpose of this step is to accurately capture more details in the damage signature area map, such as the width, depth and direction of the crack, and the size and shape of the corrosion spot. And then, generating a characteristic region map fusion rule according to the storage region damage detection frame. These rules define how the different lesion characterization area maps are combined together based on the location and size of the lesion field identified in the lesion detection frame to achieve the feature enhancement. For example, if a crack is found in one corner of the memory stripe and corrosion is found in the other corner, the fusion rules will instruct how to combine these different damage signature area maps so that both types of damage are highlighted simultaneously in the final image. And fusing the target feature enhanced images through an image fusion layer to generate a final target damage region enhanced image. Image fusion is a process of combining multiple images into a single image, with the aim of retaining all important information in the original image, while removing redundant information. In this process, the different lesion characterization area maps are effectively combined to produce a composite image containing all of the critical lesion characterizations. For example, the features of cracks and corrosion spots will be simultaneously clearly visible in the final enhanced image, providing comprehensive visual information for further damage assessment and analysis.

In a specific embodiment, the process of executing step S105 may specifically include the following steps:

(1) Marking the damage component information of the target damage region enhanced image through a damage marking network in the initial memory stripe damage detection model to obtain a target damage component mask;

(2) Performing local damage optimization on the target damage region enhanced image according to the target damage component mask to obtain a target local optimized image;

(3) And outputting a result of the target local optimization image to obtain a memory bank damage detection result.

Specifically, the target damage region enhanced image is processed through the damage labeling network, and various damage characteristics in the image, such as cracks, scratches or abrasion, are identified. This may be achieved by image recognition algorithms, such as pattern recognition techniques based on convolutional neural networks. The network accurately identifies the location, shape and size of the lesion by analyzing pixel distribution, color change and texture features in the image. The network then marks these identified lesion areas to form a target lesion assembly mask. This mask is a layer that is overlaid on the original image, visually showing the specific location and extent of the lesion. For example, if a crack is found in a certain area of the memory stripe, the mask will accurately cover that area, highlighting the location and morphology of the crack. And (3) carrying out local damage optimization on the target damage region enhanced image according to the target damage component mask, and further improving the visual effect of the damage region so that the damage region is clearer and more obvious. Details and visibility of the damaged area are enhanced by various image processing techniques such as local contrast enhancement, sharpening or color correction. The local optimization treatment not only makes the damage characteristics more striking, but also is helpful for subsequent damage assessment and repair decisions. For example, a small crack can be more easily detected and evaluated by a local optimization process, thereby providing an accurate reference for maintenance work. And outputting a result of the target local optimization image to obtain a final memory bank damage detection result. The processed image and related damage information are integrated together to form a detailed detection report. The report contains not only a visual representation of the lesion, but also information about the type, size, location, and cause of the lesion. For example, for a crack on a memory stick, the report will show a detailed image of the crack, identify the specific location of the crack, describe the length and width of the crack, and provide an analysis as to the cause of crack formation.

In a specific embodiment, the process of executing step S106 may specifically include the following steps:

(1) Performing result feedback on the memory stripe damage detection result based on a preset intelligent feedback mechanism to obtain target feedback information;

(2) Performing model parameter range analysis on the initial memory bank damage detection model according to the target feedback information to obtain a model parameter range set;

(3) Generating a plurality of model parameters of the initial memory bank damage detection model through a model parameter range set to obtain a corresponding random initial value set, and constructing a particle population from the random initial value set through a preset particle swarm optimization algorithm to obtain the particle population;

(4) Carrying out particle fitness calculation on the particle population to obtain a particle fitness set corresponding to the particle population, and carrying out iterative calculation and optimization solution on the particle fitness set to obtain a network parameter optimization set;

(5) And updating and optimizing network parameters of the initial memory bank damage detection model according to the network parameter optimization set to obtain a target memory bank damage detection model.

Specifically, the result feedback is performed on the memory bank damage detection result based on a preset intelligent feedback mechanism so as to acquire target feedback information. This mechanism evaluates the performance of the current model by analyzing data in the detection results, such as the recognition accuracy, omission rate, false detection rate, etc. of the damage. The feedback information includes not only the performance of the model on the specific lesion type, but also the influence of environmental factors on the detection result, such as illumination conditions, image quality, etc. For example, if the recognition accuracy of the model is low in low light conditions, the feedback information will indicate this and provide direction for subsequent model optimization. And carrying out model parameter range analysis on the initial memory bank damage detection model according to the target feedback information so as to obtain a model parameter range set. The feedback information is analyzed to determine which model parameters need to be adjusted, and the reasonable adjustment range of the parameters. The determination of the range of model parameters is based on a deep understanding of factors affecting the model performance, including the sensitivity of the parameters to lesion recognition, interactions between the parameters, and the like. For example, if the feedback information indicates that the model is inaccurate for certain types of crack identification, then parameters associated with edge detection need to be adjusted. And generating random initial values of a plurality of model parameters of the initial memory bank damage detection model through the model parameter range set to obtain a corresponding random initial value set. The parameter values are randomly selected within the parameter range to provide an initial set of parameters for the application of the optimization algorithm. The generation of random initial values ensures the diversity and breadth of the optimized search. For example, for the selection of the convolution kernel size or activation function in a convolutional neural network, multiple sets of initial values may be randomly generated over a range. And carrying out particle population construction on the random initial value set through a preset particle population optimization algorithm to obtain a particle population. The particle swarm optimization algorithm is an optimization algorithm that simulates the behavior of a shoal or shoal of fish, and finds the optimal solution by simulating the search behavior of particles (i.e., candidate solutions) in the solution space. Each particle represents a set of model parameters, and the position and velocity of the particle is dynamically adjusted based on the historical behavior of the particle and the behavior of other particles in the population. For example, if a combination of parameters represented by a certain particle performs well in damage detection, that particle will be considered a "leader" and the other particles will move in this direction. And carrying out particle fitness calculation on the particle population to obtain a particle fitness set corresponding to the particle population, and carrying out iterative calculation and optimization solution on the particle fitness set. Particle fitness calculation is evaluated based on the performance of the parameter combinations represented by the particles in the lesion detection task, with higher fitness meaning better parameter combinations. Through iterative computation, each particle in the particle population continuously updates its position and speed according to the performances of the particle population and the particle population, so that the optimal solution is gradually approximated. And updating and optimizing network parameters of the initial memory bank damage detection model according to the network parameter optimization set to obtain a target memory bank damage detection model. And applying the optimal parameter combination obtained by the particle swarm optimization algorithm to the initial model, thereby generating a final optimization model. The optimized model is finer in parameter setting, and damage on the memory bank can be accurately identified and analyzed.

The method for quickly detecting the memory bank damage based on the machine vision in the embodiment of the present application is described above, and the system for quickly detecting the memory bank damage based on the machine vision in the embodiment of the present application is described below, referring to fig. 2, and one embodiment of the system for quickly detecting the memory bank damage based on the machine vision in the embodiment of the present application includes:

the detection module 201 is configured to perform image detection on a target memory bank to obtain an initial memory bank image, and perform storage region feature extraction on the initial memory bank image to obtain a first storage region feature image;

the computing module 202 is configured to perform detection environment analysis on the initial memory bank image to obtain detection environment parameter data, and perform storage area feature detection weight computation on the first storage area feature image according to the detection environment parameter data to obtain a second storage area feature image;

The input module 203 is configured to input the second storage area feature image into a preset initial memory bank damage detection model, and perform storage area damage detection through a storage area damage detection network in the initial memory bank damage detection model, so as to obtain a storage area damage detection frame;

The decomposition module 204 is configured to perform feature decomposition on the second storage area feature image according to the storage area damage detection frame through a damage feature extraction network in the initial memory bank damage detection model to obtain S first damage feature area images and K second damage feature area images, and perform damage area feature enhancement and image fusion on the S first damage feature area images and the K second damage feature area images, so as to generate a target damage area enhanced image, where S and K are positive integers, and s=3k;

the labeling module 205 is configured to label and output a result of the damage component information on the target damage region enhanced image through a damage labeling network in the initial memory bank damage detection model, so as to obtain a memory bank damage detection result;

and the optimization module 206 is configured to perform network parameter optimization on the initial memory bank damage detection model according to the memory bank damage detection result by using a particle swarm optimization algorithm to obtain a network parameter optimization set, and perform network parameter update and optimization on the initial memory bank damage detection model according to the network parameter optimization set to obtain a target memory bank damage detection model.

And carrying out image detection and storage area feature extraction on the memory strip through the cooperative cooperation of the components, so that a focus of damage detection is placed on a key area. The localization method greatly improves the detection efficiency and reduces the waste of computing resources, thereby completing the memory bank damage detection more rapidly. By detecting environment analysis and storing area characteristic detection weight calculation, the method can adapt to different detection environments and conditions. The memory bank damage detection method has the advantages that the memory bank damage detection is more robust, the high efficiency can be maintained under various illumination, angles and environments, and false alarms and missing alarms are reduced. And adopting the feature decomposition and the image fusion of the damage assembly to realize the multi-level analysis of the damage of the memory bank. The method is helpful for more accurately identifying and positioning different types of damage, so that the detection result is more detailed, and more targeted information is provided for subsequent maintenance or replacement. By adopting the particle swarm optimization algorithm, the network parameters of the memory bank damage detection model can be automatically optimized and optimized to adapt to different damage types and environmental conditions, so that the performance and generalization capability of the detection model are improved. Through the damage labeling network, not only can the damage of the memory bank be detected, but also the damage assembly can be labeled and analyzed in detail, so that the accuracy of the rapid detection of the damage of the memory bank is improved.

It will be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, systems and units may refer to the corresponding processes in the foregoing method embodiments, which are not repeated herein.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a random access memory (random acceS memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The above embodiments are only for illustrating the technical solution of the present application, and not for limiting the same; although the application has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. The quick memory bank damage detection method based on machine vision is characterized by comprising the following steps of:

Performing detection environment analysis on the initial memory bank image to obtain detection environment parameter data, and performing storage area feature detection weight calculation on the first storage area feature image according to the detection environment parameter data to obtain a second storage area feature image; the method specifically comprises the following steps: detecting environment analysis is carried out on the initial memory bank image to obtain initial environment parameter data, and a loss detection environment influence factor set is obtained; performing environmental parameter analysis on the initial environmental parameter data according to the loss detection environmental impact factor set to obtain detection environmental parameter data; performing environmental impact factor hierarchical analysis on the detection environment parameter data through a preset hierarchical analysis method, and constructing a hierarchical structure, wherein the hierarchical structure comprises a target layer, a criterion layer and a factor layer; constructing a discrimination matrix of the detection environment parameter data according to the hierarchical structure, and carrying out mean value operation on a plurality of column vectors in the discrimination matrix to obtain a weight vector; carrying out normalization processing on the weight vector to obtain storage area feature detection weight data, and carrying out storage area feature detection weight calculation on the first storage area feature image according to the storage area feature detection weight data to obtain a second storage area feature image;

2. The method for quickly detecting memory bank damage based on machine vision according to claim 1, wherein the performing image detection on the target memory bank to obtain an initial memory bank image, and performing storage region feature extraction on the initial memory bank image to obtain a first storage region feature image comprises:

Performing image detection on a target memory bank to obtain an initial memory bank image, performing noise removal on the initial memory bank image to obtain a first memory bank image, and performing edge enhancement on the first memory bank image to obtain a second memory bank image;

Performing brightness parameter correction on the second memory bank image to obtain brightness correction parameter data, performing color space conversion on the second memory bank image to obtain color conversion parameter data, and performing perception weight distribution on the second memory bank image to obtain perception weight parameter data;

Obtaining a memory bank characteristic parameter data set according to the brightness correction parameter data, the color conversion parameter data and the perception weight parameter data;

And carrying out storage area parameter configuration and storage area feature extraction on the second memory bank image according to the memory bank feature parameter data set to obtain a first storage area feature image.

3. The machine vision-based memory bank damage rapid detection method according to claim 1, wherein the inputting the second memory bank feature image into a preset initial memory bank damage detection model, and performing memory bank damage detection through a memory bank damage detection network in the initial memory bank damage detection model, to obtain a memory bank damage detection frame, includes:

Inputting the second storage area characteristic image into a storage area damage detection network in a preset initial memory stripe damage detection model, wherein the storage area damage detection network comprises an image standardization layer, a storage area damage abnormality detection layer and an anchor frame layer;

Performing image size standardization processing on the second storage area characteristic image through the image standardization layer to obtain a standard storage area characteristic image;

Performing storage area damage detection on the standard storage area characteristic image through a two-layer convolution network in the storage area damage abnormal detection layer to obtain a probability value that each pixel belongs to a damage abnormal area;

and generating a corresponding storage area damage abnormal probability map according to the probability value that each pixel belongs to the damage abnormal area, and judging the damage abnormal area and generating a detection frame of the storage area damage abnormal probability map according to a preset probability threshold value to obtain a storage area damage detection frame.

4. The machine vision-based memory bank damage rapid detection method according to claim 3, wherein the damage feature extraction network in the initial memory bank damage detection model performs damage component feature decomposition on the second storage area feature image according to the storage area damage detection frame to obtain S first damage feature area diagrams and K second damage feature area diagrams, performs damage area feature enhancement and image fusion on the S first damage feature area diagrams and the K second damage feature area diagrams, and generates a target damage area enhanced image, wherein S and K are positive integers, s=3k, and the method comprises:

inputting the second storage area characteristic image into a damage characteristic extraction network in the initial memory bank damage detection model, wherein the damage characteristic extraction network comprises a damage component characteristic decomposition layer, an improved Laplace operator and an image fusion layer;

Decomposing the second storage area characteristic image into S first damage characteristic area diagrams and K second damage characteristic area diagrams through the damage component characteristic decomposition layer;

Capturing damage characteristic region details of the S first damage characteristic region graphs and the K second damage characteristic region graphs through the improved Laplacian operator;

Generating a characteristic region map fusion rule according to the storage region damage detection frame, and carrying out damage region feature enhancement on the S first damage characteristic region maps and the K second damage characteristic region maps according to the characteristic region map fusion rule to obtain a plurality of target feature enhancement images, wherein S and K are positive integers, and S=3K;

And carrying out image fusion on the plurality of target feature enhanced images through the image fusion layer to generate a target damage region enhanced image.

5. The method for quickly detecting memory bank damage based on machine vision according to claim 4, wherein the performing damage component information labeling and result output on the target damage region enhanced image through the damage labeling network in the initial memory bank damage detection model to obtain a memory bank damage detection result comprises:

Marking the damage component information of the target damage region enhanced image through a damage marking network in the initial memory bank damage detection model to obtain a target damage component mask;

performing local damage optimization on the target damage region enhanced image according to the target damage component mask to obtain a target local optimized image;

and outputting a result of the target local optimization image to obtain a memory bank damage detection result.

6. The machine vision-based memory bank damage rapid detection method according to claim 1, wherein the performing network parameter tuning on the initial memory bank damage detection model according to the memory bank damage detection result by using a particle swarm optimization algorithm to obtain a network parameter optimization set, and performing network parameter updating and optimization on the initial memory bank damage detection model according to the network parameter optimization set to obtain a target memory bank damage detection model comprises:

performing result feedback on the memory bank damage detection result based on a preset intelligent feedback mechanism to obtain target feedback information;

Performing model parameter range analysis on the initial memory bank damage detection model according to the target feedback information to obtain a model parameter range set;

Generating a plurality of model parameters of the initial memory bank damage detection model through the model parameter range set to obtain a corresponding random initial value set, and constructing a particle population to the random initial value set through a preset particle population optimization algorithm to obtain the particle population;

Performing particle fitness calculation on the particle population to obtain a particle fitness set corresponding to the particle population, and performing iterative calculation and optimization solution on the particle fitness set to obtain a network parameter optimization set;

And updating and optimizing network parameters of the initial memory bank damage detection model according to the network parameter optimization set to obtain a target memory bank damage detection model.

7. The utility model provides a memory bank damage short-term test system based on machine vision, its characterized in that, memory bank damage short-term test system based on machine vision includes:

The computing module is used for carrying out detection environment analysis on the initial memory bank image to obtain detection environment parameter data, and carrying out storage area feature detection weight computation on the first storage area feature image according to the detection environment parameter data to obtain a second storage area feature image; the method specifically comprises the following steps: detecting environment analysis is carried out on the initial memory bank image to obtain initial environment parameter data, and a loss detection environment influence factor set is obtained; performing environmental parameter analysis on the initial environmental parameter data according to the loss detection environmental impact factor set to obtain detection environmental parameter data; performing environmental impact factor hierarchical analysis on the detection environment parameter data through a preset hierarchical analysis method, and constructing a hierarchical structure, wherein the hierarchical structure comprises a target layer, a criterion layer and a factor layer; constructing a discrimination matrix of the detection environment parameter data according to the hierarchical structure, and carrying out mean value operation on a plurality of column vectors in the discrimination matrix to obtain a weight vector; carrying out normalization processing on the weight vector to obtain storage area feature detection weight data, and carrying out storage area feature detection weight calculation on the first storage area feature image according to the storage area feature detection weight data to obtain a second storage area feature image;