CN116664573B - Downhole drill rod number statistics method based on improved YOLOX - Google Patents
Downhole drill rod number statistics method based on improved YOLOX Download PDFInfo
- Publication number
- CN116664573B CN116664573B CN202310944258.6A CN202310944258A CN116664573B CN 116664573 B CN116664573 B CN 116664573B CN 202310944258 A CN202310944258 A CN 202310944258A CN 116664573 B CN116664573 B CN 116664573B
- Authority
- CN
- China
- Prior art keywords
- feature
- image
- counting
- frame
- drill rod
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000000605 extraction Methods 0.000 claims abstract description 21
- 238000001514 detection method Methods 0.000 claims abstract description 15
- 230000007246 mechanism Effects 0.000 claims abstract description 13
- 238000005553 drilling Methods 0.000 claims abstract description 10
- 238000007781 pre-processing Methods 0.000 claims abstract description 9
- 238000005457 optimization Methods 0.000 claims abstract description 5
- 238000012544 monitoring process Methods 0.000 claims abstract description 4
- 230000006870 function Effects 0.000 claims description 35
- 238000001914 filtration Methods 0.000 claims description 12
- 238000011176 pooling Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 8
- 230000002457 bidirectional effect Effects 0.000 claims description 7
- 238000010586 diagram Methods 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 6
- 230000006872 improvement Effects 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 6
- 239000003245 coal Substances 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 4
- 230000008030 elimination Effects 0.000 claims description 4
- 238000003379 elimination reaction Methods 0.000 claims description 4
- 210000001503 joint Anatomy 0.000 claims description 4
- 230000002708 enhancing effect Effects 0.000 claims description 3
- 230000002401 inhibitory effect Effects 0.000 claims description 3
- 230000000737 periodic effect Effects 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 230000003213 activating effect Effects 0.000 claims description 2
- 238000002474 experimental method Methods 0.000 claims description 2
- 238000005070 sampling Methods 0.000 claims description 2
- 238000005728 strengthening Methods 0.000 claims description 2
- 239000013598 vector Substances 0.000 claims description 2
- 230000001105 regulatory effect Effects 0.000 claims 1
- 238000005065 mining Methods 0.000 abstract description 4
- 238000005516 engineering process Methods 0.000 abstract description 2
- 239000011435 rock Substances 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 239000012535 impurity Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003032 molecular docking Methods 0.000 description 1
- 238000012946 outsourcing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration using local operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/40—Image enhancement or restoration using histogram techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
- G06T2207/20032—Median filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30204—Marker
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30242—Counting objects in image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Multimedia (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Quality & Reliability (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an improved YOLOX-based underground drill rod number counting method, which belongs to the field of mining and information technology, and comprises the following steps: reading a monitoring video of underground drilling, and preprocessing an image acquired by frame extraction through frame extraction; introducing a CBAM attention mechanism to improve the model feature extraction capability; using a BiFPN feature pyramid to reduce feature loss; the optimization loss function is CIOU, and the positioning capability of the model prediction frame is improved; performing target detection and tracking by using an improved model trained by the preprocessed image; and eliminating small peaks without the unloading operation by setting a peak threshold value, recording the rest effective peaks, and counting once if one peak exists. The invention improves the counting accuracy and has important significance for automatically counting the number of the drill rods.
Description
Technical Field
The invention relates to the field of mining and information technology, in particular to an underground drill rod number counting method based on improved YOLOX.
Background
Rock burst is one of the most serious disasters in deep coal mining in China, and compared with other methods for preventing and controlling rock burst, the method for drilling and releasing pressure has the advantages of simple process, low cost and the like, and is widely applied to coal mining. Because the pressure relief drilling depth is difficult to directly measure, the pressure relief drilling depth is generally calculated indirectly through the number of drill rods. In practical measurement, drill rod specifications are the same in the same drilling operation, so drill rod numbers are generally used for measuring drilling depths. The traditional drill rod counting method is manual counting, the method is low in efficiency, and the insufficient pressure relief drilling depth possibly occurs due to false alarm risks in outsourcing engineering, so that potential safety hazards are brought to coal mine production.
Disclosure of Invention
Aiming at the technical problems in the prior art, the invention provides a downhole drill rod number counting method based on improved YOLOX (You Only Look Once: unified, real-Time Object Detection X, only once: unified Real-time object detection X version), which is reasonable in design, overcomes the defects of the prior art and has a good effect.
In order to achieve the above purpose, the present invention adopts the following technical scheme: a downhole drill pipe quantity statistics method based on improved YOLOX, comprising the steps of: step 1: preprocessing a frame extraction image; reading a monitoring video of underground drilling, performing frame extraction, and preprocessing an image obtained by the frame extraction to obtain an original data set; step 2: improvement of training model; introducing CBAM (Convolutional Block Attention Module, convolution attention module) into the model to improve the model feature extraction capability; using bippn (Bidirectional Feature Pyramid Network, bi-directional feature pyramid network) to reduce feature loss; the optimization loss function is CIOU (Complete Intersection-over-Union), and the positioning capability of the model prediction frame is improved; step 3: target detection and tracking; performing target detection and tracking by using an improved model trained by the preprocessed images, wherein the detection targets are a power head and the front end of a machine body, and drawing a periodic interval change waveform chart of the power head and the front end of the machine body; step 4: wave crest screening and counting; and eliminating small peaks without the unloading operation by setting a peak threshold value, recording the rest effective peaks, and counting once if one peak exists.
Preferably, the step 1 specifically includes the following steps: step 1.1: image denoising; carrying out noise elimination and filtering treatment on the frame extraction image by a median filtering method; step 1.2: enhancing the image; the image after noise cancellation and filtering is enhanced through histogram equalization.
Preferably, the step 2 specifically includes the following steps: step 2.1: an attention introducing mechanism; introducing CBAM into the model, automatically acquiring the importance degree of the feature channel through the CBAM, giving weight, strengthening important features and inhibiting non-important features; step 2.2: introducing a feature pyramid; the feature pyramid realizes the fusion of high-level semantic information and low-level detail information through a bidirectional channel from top to bottom and from bottom to top; step 2.3: optimizing the loss function to be CIOU; the optimal loss function is a CIOU loss function, gradient return is realized through the CIOU loss function, positioning of the prediction frame is facilitated, and the similarity degree of the aspect ratio of the prediction frame and the real frame is reflected.
Preferably, the step 3 specifically includes the following steps: step 3.1: predicting the front ends of the power head and the machine body by using the trained model to obtain prediction frames of the power head and the machine body; step 3.2: calculating the distance between the two prediction frame center points; step 3.3: 20 frames of images are extracted every minute, a pitch waveform is drawn with the calculated pitch, the horizontal axis is time, and the vertical axis is distance between the two.
Preferably, the step 4 specifically includes the following steps: step 4.1: obtaining peak value and time information at a peak according to the information of the waveform diagram; step 4.2: because the butt joint power head and the drill rod are in butt joint, small wave peaks without the operation of unloading the drill rod appear, the wave peak threshold value is set for the small wave peaks to be removed; step 4.3: and counting the wave crests meeting the threshold requirement, and carrying out one-time rod unloading operation at the wave crests at the moment, wherein one drill rod is unloaded in one-time rod unloading operation, so that the wave crest count is the statistical quantity of the drill rods.
The invention has the beneficial technical effects that: (1) The method can weaken the interference caused by underground dimming conditions on identification by the image preprocessing method, and has better universality on more accurate identification of objects under the channel; (2) The recognition accuracy reaches 91.3% by improving the model; the more accurate prediction of the object under the underground complex background can be realized; (3) Small peaks which do not generate rod unloading operation are filtered through setting of the peak threshold value, and statistics of the number of drill rods is achieved through recording of the number of effective peaks.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
The invention is described in further detail below with reference to the attached drawings and detailed description: a downhole drill rod number counting method based on improved YOLOX, which is shown in fig. 1, and comprises the following steps: step 1: reading a monitoring video of underground drilling, preprocessing an image acquired by frame extraction through frame extraction, and solving the problems of low noise and low contrast in the underground image; the method specifically comprises the following steps: step 1.1: image filtering and denoising; and smoothing the images in the data set through a filter in sequence to generate new images to form a new data set, and inhibiting noise of the target image under the condition of reserving the detail characteristics of the images as much as possible so as to eliminate the problem of image noise caused by excessively high equipment sensitivity setting caused by the underground dimming environment.
Taking out frames of video data in a certain coal mine rod unloading process, preprocessing an image after taking out frames to be used as an original data set, wherein the data set has 4783 pictures in total; carrying out noise elimination treatment on the frame-extracted image by different filtering methods, and selecting a more proper noise elimination method; and respectively adopting median filtering, mean filtering and Gaussian filtering to carry out transverse comparison, and finally selecting the median filtering which can eliminate image noise and has the smallest influence on image definition according to comparison results.
Step 1.2: enhancing the image; because the whole underground environment is dim, the gray level of the image data is concentrated and distributed at a lower level, the contrast of the image is insufficient, and the image is required to be enhanced. The filtered image is enhanced by histogram equalization. The contrast of the key area of the image after the global histogram equalization treatment is insufficient, the gray level of the image after the treatment is still concentrated at a lower level, so that the equalized image is marked by adopting the local histogram equalization, the marking targets are two parts of a power head and the front end of a machine body, and the data set is divided into a training set and a test set according to the proportion of 8:2.
The input picture is firstly converted into an HSI color space, histogram equalization is carried out on the intensity channel, and then the processed result is converted into an RGB color space. The mapping of each pixel from the original gray level to the new gray level is realized, the gray level conversion is realized, and the contrast of the image is improved.
Step 2: model improvement. Three-point improvement is provided for the YOLOX network framework according to the underground environment conditions, so that the model is helped to achieve better feature extraction and fusion, and the model is positioned more accurately. Introducing a CBAM attention mechanism into the model to improve the feature extraction capability of the model; using a BiFPN feature pyramid to reduce feature loss; and the optimization loss function is CIOU, so that the positioning capability of the model prediction frame is improved.
The method specifically comprises the following steps: step 2.1: fusing a attentive mechanism; insufficient underground illumination, poor contrast, and poor positioning and recognition accuracy can be caused by fuzzy boundary between a target object and a background due to impurities, pipelines, equipment and the like. The attention mechanism can automatically acquire the importance degree of the characteristic channel and give weight to strengthen important characteristics and inhibit non-important characteristics. The CBAM attention mechanism is fused in the backbone network (Convolutional Block Attention Module).
Firstly, compressing a feature map F in a space dimension through a channel attention mechanism, respectively carrying out global average pooling and maximum pooling on the height and width, and then passing through a perceptron model MLP; then the feature elements are summed and then subjected to inner product operation with the original feature image through a Sigmoid activation functionRepresenting the inner product and outputting a characteristic diagram F 1 As an input profile for the spatial attention module. The spatial attention mechanism also carries out global average pooling and maximum pooling processes, then carries out convolution operation through a convolution kernel, and carries out inner product operation on the input and output characteristic diagrams after activating functions, wherein the formula is as follows.
The channel attention formula and spatial attention mechanism are shown below.
Wherein M is C (F) And M S (F) Channel and spatial attention vectors, respectively, σ represents the activation function, here Sigmoid function, avgp and Maxp represent the average pooling and maximum pooling operations, respectively, f 7×7 Representing a convolution operation with a convolution kernel size of 7 x 7.
Step 2.2: introducing a feature pyramid; the feature pyramid realizes the fusion of high-level semantic information and low-level detail information through a bidirectional channel from top to bottom and from bottom to top; the front end of the power head and the front end of the machine body belong to small targets in the recognition task, the space dimension of input data can be reduced when a convolution kernel slides on the input data, the size of a feature map can be reduced, extraction of higher-level features is facilitated, and the loss of feature information of small objects can be caused in the process. The feature pyramid fuses the up-sampling of the deep features with the shallow features to achieve a prediction effect, so that the loss of small target feature information can be reduced. BiFPN (Bidirectional Feature Pyramid Network) fuses the information of different scales through the two-way channels from top to bottom and from bottom to top, sets corresponding weights according to the importance of the feature information, can effectively adjust the contribution degree of the feature information in the feature map, and realizes better feature fusion.
Step 2.3: optimizing a loss function; the original IOU loss function of the YOLO is optimized to be the CIOU loss function.
A dim downhole environment may result in poor positioning of the prediction frame. The loss function can reflect the degree of difference between the predicted frame and the real frame and optimize the positioning of the predicted frame through gradient return. But the different loss functions have different capacities of reflecting the difference between the prediction frame and the real frame, and good loss functions can better reflect the difference condition and realize better echelon return. The positioning of the prediction frame can be more in line with the real frame, and the calculation of the center point distance of the follow-up prediction frame is more accurate.
CIOU (Complete IoU), gradient feedback can be realized when the frame bodies are not overlapped or are all contained, positioning of the prediction frame is facilitated, and the similarity degree of the length-width consistency parameter between the prediction frame and the real frame can be reflected.
Wherein IOU is the intersection ratio of the areas of the predicted frame and the real frame, A is the predicted frame, B is the center point coordinate of the predicted frame, B is the real frame gt Is the coordinate ρ of the center point of the real frame 2 (b,b gr ) Representing the square of the distance between two center points, c 2 The square of the minimum bounding rectangle diagonal length of the predicted and real frames is represented, K is a parameter for track-off, and η is a parameter for measuring the consistency of the aspect ratio. The CIOU loss function can better reflect the difference between the prediction frame and the real frame, realize better gradient return, help the prediction frame to realize more accurate positioning, and reduce the influence of the underground dim environment on the prediction positioning.
Step 3: training a model; performing target detection and tracking by using an improved model trained by the preprocessed images, wherein the detection targets are a power head and the front end of a machine body, and drawing a periodic interval change waveform chart of the power head and the front end of the machine body; we run on a computer with Windows10 operating system, geForce RTX 3060, indelphinium, and Pytorch deep learning framework using python3.7 as the programming language. The specific experimental initial parameter settings are shown in table 1.
TABLE 1
The invention selects YOLOV4, V5, SSD, YOLOX and improved model for comparison study. mAP and FPS values on the test set are shown in Table 2.
TABLE 2
As can be seen from table 2, the YOLOX performance before the improvement is also better than the previous generation version. The SSD algorithm detects at a much higher speed than the rest of the models, but its detection accuracy is also lowest. After the improvement, the mAP of the YOLOX model reaches 91.3%, and is respectively improved by 15.1% and 11.5% compared with V4 and V5, and the FPS value is slightly improved. The mAP value was increased by 4.4% with less FPS value reduction compared to the original model. The improved model ensures the detection speed while greatly improving the detection precision.
And (5) waveform drawing. The improved model realizes the accurate positioning of the power head and the front end of the machine body through training, and the central position of a prediction frame of the power head and the front end of the machine body at t can be obtained and the distance S between the power head and the front end of the machine body can be calculated according to a distance formula t The formula is as follows:wherein S is t At t, the distance between the power head and the front end of the machine body (x) 2t ,y 2t ),(x 1t ,y 1t ) And the central coordinates of the prediction frames of the power head and the front end of the machine body are respectively t. The spacing was calculated at 20 frames per minute uniformly to plot the spacing curve.
Step 4: wave crest screening and counting; and eliminating small peaks without the unloading operation by setting a peak threshold value, recording the rest effective peaks, and counting once if one peak exists.
The rod unloading is a process that the power head is close to the drill rod and connected with the drill rod and then drives the drill rod to move until the drill rod is taken out completely and then manually unloaded. Because the shape and other characteristics of the drill rod are not obvious, the drill rod is difficult to capture in an actual detection task, and the difficulty of directly identifying the position of the drill rod is high. Considering that the relative position of the power head and the front end of the machine body has an indirect relation with the unloading operation, the counting of the unloading drill rods can be realized through the change condition of the distance relation between the power head and the front end of the machine body.
A large peak close to 1m of the length of a single drill rod appears in the curve, and one drill rod is removed corresponding to one-time rod removing operation. However, small peaks appear in the same curve, and the intervals of the small peaks do not meet the requirement of taking out all drill rods, so that the rod unloading operation does not occur. In combination with video analysis, the reason is that the position needs to be fine-tuned for docking when the power head is docked with the drill pipe. At this time, the power head is close to the front end of the machine body, and the position of the power head is not changed greatly, so that the height of the small wave peak is far lower than the height of the wave peak when the lever unloading operation occurs. The large peak is removed from a drill rod and should count to 1, and the small peak is not removed from the drill rod and does not count. Therefore, a crest threshold can be set, the crest meeting the threshold requirement is a valid crest and is counted, otherwise, the crest is not counted to remove small crest interference.
Considering factors such as underground image quality, recognition errors and the like, a crest of the length of the drill rod which is tentatively set to be 0.85 times is an effective crest and recorded, a crest validity judging function omega is set, when the crest height exceeds a threshold value, the crest is recorded as 1, and otherwise, the crest height is recorded as 0.
Wherein PKV is n A peak value representing the nth peak, L Rod The length of the drill rods was 1m in this experiment for the length of a single drill rod, so the spacing threshold was set at 85cm. When the value of ω is 1, it means that the threshold requirement is satisfied; and when the value is 0, it is not satisfied.
A tripping operation and tripping out of a drill rod occurs at the peak meeting the threshold requirement, and a count is made.
The formula is as follows:wherein PKV represents a set of peaks and D represents S t The findpeaks is a peak-solving function, the numerical value of the wave crest can be recorded, countif is a conditional counting function, when ω=1, the function counts once, otherwise, the function does not count, and the final counting result is output to be the counting result of the number of the drill rods.
It should be understood that the above description is not intended to limit the invention to the particular embodiments disclosed, but to limit the invention to the particular embodiments disclosed, and that the invention is not limited to the particular embodiments disclosed, but is intended to cover modifications, adaptations, additions and alternatives falling within the spirit and scope of the invention.
Claims (1)
1. The underground drill rod number counting method based on the improved YOLOX is characterized by comprising the following steps of: the method comprises the following steps:
step 1: preprocessing a frame extraction image;
reading a monitoring video of underground drilling, performing frame extraction, and preprocessing an image obtained by the frame extraction to obtain an original data set;
step 2: improvement of training model;
introducing CBAM to the model to improve the feature extraction capability of the model; using BiFPN to reduce feature loss; the optimization loss function is CIOU, and the positioning capability of the model prediction frame is improved;
step 3: target detection and tracking;
performing target detection and tracking by using an improved model trained by the preprocessed images, wherein the detection targets are a power head and the front end of a machine body, and drawing a periodic interval change waveform chart of the power head and the front end of the machine body;
step 4: wave crest screening and counting;
removing small wave peaks without lever unloading operation by setting a wave peak threshold value, recording the rest effective wave peaks, and counting once if one wave peak exists;
the step 1 specifically comprises the following steps:
step 1.1: image denoising;
the method comprises the steps of performing frame extraction on video data in a certain coal mine rod unloading process, preprocessing an image after frame extraction to obtain an original data set, performing 4783 pictures on the data set, and performing noise elimination filtering on the frame extraction image by a median filtering method;
step 1.2: enhancing the image;
the image after noise cancellation and filtering is enhanced through histogram equalization; marking the equalized image by adopting local histogram equalization, wherein the marking target is two parts of a power head and the front end of a machine body, and the data set is divided into a training set and a testing set according to the proportion of 8:2;
the step 2 specifically comprises the following steps:
step 2.1: an attention introducing mechanism;
introducing CBAM into the model, automatically acquiring the importance degree of the feature channel through the CBAM, giving weight, strengthening important features and inhibiting non-important features;
firstly, compressing a feature map F in a space dimension through a channel attention mechanism, respectively carrying out global average pooling and maximum pooling on the height and width, and then passing through a perceptron model MLP; then the feature elements are summed and then subjected to inner product operation with the original feature image through a Sigmoid activation functionRepresenting the inner product and outputting a characteristic diagram F 1 An input feature map as a spatial attention module; the space attention mechanism also carries out global average pooling and maximum pooling processes, then carries out convolution operation through a convolution kernel, and carries out inner product operation on the input and output characteristic diagrams after activating functions, wherein the formula is as follows:
the channel attention formula and spatial attention mechanism are as follows:
M C (F)=σ[MLP(Avgp(F)+Maxp(F))] (3);
M S (F)=σ(f 7×7 [Avgp(F);Maxp(F)]) (4);
wherein M is c (F) And M s (F) Channel and spatial attention vectors, respectively, σ represents the activation function, here Sigmoid function, avgp and Maxp represent the average pooling and maximum pooling operations, respectively, f 7×7 Representing a convolution operation with a convolution kernel size of 7 x 7;
step 2.2: introducing a feature pyramid;
the feature pyramid realizes the fusion of high-level semantic information and low-level detail information through a bidirectional channel from top to bottom and from bottom to top; the front end of the power head and the front end of the machine body belong to small targets in the identification task, the space dimension of input data can be reduced when a convolution kernel slides on the input data, the dimension of a feature map can be reduced, the extraction of higher-level features is facilitated, and the loss of feature information of small objects can be caused in the process; the feature pyramid fuses the up-sampling of the deep features with the shallow features to achieve a prediction effect, so that the loss of small target feature information is reduced; the BiFPN fuses information of different scales through a bidirectional channel from top to bottom and a bidirectional channel from bottom to top, and corresponding weights are set according to the importance of the feature information, so that the contribution degree of the feature information in the feature map can be effectively regulated, and better feature fusion is realized;
step 2.3: optimizing the loss function to be CIOU;
the optimization loss function is a CIOU loss function, gradient return is realized through the CIOU loss function, positioning of the prediction frame is facilitated, and the similarity degree of the aspect ratio of the prediction frame and the real frame is reflected;
the CIOU loss function can better reflect the difference between the prediction frame and the real frame, realize better gradient return, help the prediction frame to realize more accurate positioning, and reduce the influence of the underground dimming environment on the prediction positioning;
the step 3 specifically comprises the following steps:
step 3.1: predicting the front ends of the power head and the machine body by using the trained model to obtain prediction frames of the power head and the machine body;
step 3.2: calculating the distance between the two prediction frame center points;
step 3.3: extracting 20 frames of images every minute, drawing a spacing waveform chart according to the calculated spacing, wherein the horizontal axis is time, and the vertical axis is distance between the two images;
the formula is as follows:wherein S is t At t, the distance between the power head and the front end of the machine body (x) 2t ,y 2t ),(x 1t ,y 1t ) The central coordinates of the prediction frames of the power head and the front end of the machine body are respectively t;
the step 4 specifically comprises the following steps:
step 4.1: obtaining peak value and time information at a peak according to the information of the waveform diagram;
step 4.2: because the butt joint power head and the drill rod are in butt joint, small wave peaks without the operation of unloading the drill rod appear, the wave peak threshold value is set for the small wave peaks to be removed;
step 4.3: counting the wave peaks meeting the threshold requirement, and performing one-time rod unloading operation at the wave peaks at the moment, wherein one drill rod is unloaded by one-time rod unloading operation, so that the wave peak count is the statistical number of the drill rods;
considering the underground image quality and recognition error factors, temporarily setting the crest of the drill rod length which is 0.85 times as an effective crest and recording, setting the effectiveness judgment function omega of the crest, recording as 1 when the crest height exceeds a threshold value, and recording as 0 otherwise;
wherein PKV is n A peak value representing the nth peak, L Rod The length of each drill rod is 1m in the experiment, so that the distance threshold is set to be 85cm; when the value of ω is 1, it means that the threshold requirement is satisfied; and when the value is 0, thenNot satisfied;
a first rod unloading operation and a drill rod unloading operation occur at the wave crest meeting the threshold requirement, and when the counting is performed once;
the formula is as follows:
wherein PKV represents a set of peaks and D represents S t The findpeaks is a peak-solving function, the numerical value of the wave crest can be recorded, countif is a conditional counting function, when ω=1, the function counts once, otherwise, the function does not count, and the final counting result is output to be the counting result of the number of the drill rods.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310944258.6A CN116664573B (en) | 2023-07-31 | 2023-07-31 | Downhole drill rod number statistics method based on improved YOLOX |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310944258.6A CN116664573B (en) | 2023-07-31 | 2023-07-31 | Downhole drill rod number statistics method based on improved YOLOX |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116664573A CN116664573A (en) | 2023-08-29 |
CN116664573B true CN116664573B (en) | 2024-02-09 |
Family
ID=87712203
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310944258.6A Active CN116664573B (en) | 2023-07-31 | 2023-07-31 | Downhole drill rod number statistics method based on improved YOLOX |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116664573B (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113033315A (en) * | 2021-02-26 | 2021-06-25 | 江西理工大学 | Rare earth mining high-resolution image identification and positioning method |
CN114693739A (en) * | 2022-04-11 | 2022-07-01 | 中国矿业大学 | Downhole drill rod counting method and device based on visual tracking algorithm |
CN115131712A (en) * | 2022-07-12 | 2022-09-30 | 安徽理工大学 | Intelligent counting method of drill rods driven by computer vision |
CN115909121A (en) * | 2022-08-24 | 2023-04-04 | 安徽理工大学 | Mine underground drill rod counting method, system and equipment based on improved YOLOV5 and Deepsort |
CN116051953A (en) * | 2022-11-23 | 2023-05-02 | 中国铁塔股份有限公司重庆市分公司 | Small target detection method based on selectable convolution kernel network and weighted bidirectional feature pyramid |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9706185B2 (en) * | 2012-04-16 | 2017-07-11 | Canrig Drilling Technology Ltd. | Device control employing three-dimensional imaging |
-
2023
- 2023-07-31 CN CN202310944258.6A patent/CN116664573B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113033315A (en) * | 2021-02-26 | 2021-06-25 | 江西理工大学 | Rare earth mining high-resolution image identification and positioning method |
CN114693739A (en) * | 2022-04-11 | 2022-07-01 | 中国矿业大学 | Downhole drill rod counting method and device based on visual tracking algorithm |
CN115131712A (en) * | 2022-07-12 | 2022-09-30 | 安徽理工大学 | Intelligent counting method of drill rods driven by computer vision |
CN115909121A (en) * | 2022-08-24 | 2023-04-04 | 安徽理工大学 | Mine underground drill rod counting method, system and equipment based on improved YOLOV5 and Deepsort |
CN116051953A (en) * | 2022-11-23 | 2023-05-02 | 中国铁塔股份有限公司重庆市分公司 | Small target detection method based on selectable convolution kernel network and weighted bidirectional feature pyramid |
Non-Patent Citations (3)
Title |
---|
Drill Pipe Counting Method Based on Scale Space and Siamese Network;Lihong Dong等;《ACAI 20: Proceedings of the 2020 3rd International Conference on Algorithms, Computing and Artificial》;全文 * |
基于改进ResNet网络的井下钻杆计数方法;高瑞;郝乐;刘宝;文静怡;陈宇航;;工矿自动化(第10期);全文 * |
基于改进RFB和孪生网络的煤矿钻杆计数方法研究;吴鑫艺;《中国优秀硕士学位论文全文数据库 工程科技Ⅰ辑》;全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN116664573A (en) | 2023-08-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113469177B (en) | Deep learning-based drainage pipeline defect detection method and system | |
CN111080620B (en) | Road disease detection method based on deep learning | |
CN114693615A (en) | Deep learning concrete bridge crack real-time detection method based on domain adaptation | |
CN108564085B (en) | Method for automatically reading of pointer type instrument | |
CN109852748B (en) | Method for monitoring development process of gas flow in distribution period of blast furnace and predicting gas utilization rate | |
WO2023082418A1 (en) | Power utility tunnel settlement crack identification method based on artificial intelligence technology | |
CN109376740A (en) | A kind of water gauge reading detection method based on video | |
CN110569755B (en) | Intelligent accumulated water detection method based on video | |
CN113159061B (en) | Actual tunnel surrounding rock fragment identification method based on example segmentation | |
CN110046570B (en) | Method and device for dynamically supervising grain stock of granary | |
CN112861912A (en) | Deep learning-based method and system for identifying indicator diagram of complex working condition of pumping well | |
CN114926407A (en) | Steel surface defect detection system based on deep learning | |
CN115797354A (en) | Method for detecting appearance defects of laser welding seam | |
CN115131712B (en) | Intelligent counting method for drill rod driven by computer vision | |
CN113298181A (en) | Underground pipeline abnormal target identification method and system based on dense connection Yolov3 network | |
CN109859199A (en) | A kind of method of the fresh water pipless pearl quality testing of SD-OCT image | |
CN115631191A (en) | Coal blockage detection algorithm based on gray level features and edge detection | |
CN116664573B (en) | Downhole drill rod number statistics method based on improved YOLOX | |
CN112561885B (en) | YOLOv 4-tiny-based gate valve opening detection method | |
CN110751687B (en) | Apple size grading method based on computer vision minimum and maximum circle | |
CN112597923A (en) | Pulse pile-up correction method based on morphology and optimized gray model | |
CN116740125A (en) | Detection method for realizing visualization of transparent fluid based on light refraction and deep learning | |
CN114596262A (en) | Dam monitoring and analyzing method and system based on image recognition technology | |
CN115512299A (en) | Flood early warning method of U-net variant neural network based on radar image | |
Tingjiang et al. | Research and application of downhole drilling depth based on computer vision technique |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |