WO2022213465A1 - Procédé et appareil de reconnaissance d'image à base de réseau neuronal, dispositif électronique et support - Google Patents
Procédé et appareil de reconnaissance d'image à base de réseau neuronal, dispositif électronique et support Download PDFInfo
- Publication number
- WO2022213465A1 WO2022213465A1 PCT/CN2021/097073 CN2021097073W WO2022213465A1 WO 2022213465 A1 WO2022213465 A1 WO 2022213465A1 CN 2021097073 W CN2021097073 W CN 2021097073W WO 2022213465 A1 WO2022213465 A1 WO 2022213465A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- image
- target object
- loss function
- target
- trained
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 33
- 230000006870 function Effects 0.000 claims abstract description 231
- 238000001514 detection method Methods 0.000 claims abstract description 156
- 238000003062 neural network model Methods 0.000 claims abstract description 51
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 46
- 230000008447 perception Effects 0.000 claims abstract description 21
- 230000004913 activation Effects 0.000 claims description 38
- 238000012545 processing Methods 0.000 claims description 32
- 238000012549 training Methods 0.000 claims description 30
- 238000009795 derivation Methods 0.000 claims description 26
- 238000013139 quantization Methods 0.000 claims description 20
- 238000004364 calculation method Methods 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 17
- 238000004590 computer program Methods 0.000 claims description 14
- 238000000605 extraction Methods 0.000 claims description 10
- 238000013507 mapping Methods 0.000 claims description 5
- 230000001960 triggered effect Effects 0.000 claims description 4
- 206010028980 Neoplasm Diseases 0.000 description 17
- 238000007726 management method Methods 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000008676 import Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000010191 image analysis Methods 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 210000004881 tumor cell Anatomy 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000000802 evaporation-induced self-assembly Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- 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
- G06V2201/03—Recognition of patterns in medical or anatomical images
- G06V2201/032—Recognition of patterns in medical or anatomical images of protuberances, polyps nodules, etc.
Definitions
- the present application relates to the technical field of image detection, and in particular, to a neural network-based image recognition method, apparatus, electronic device, and computer-readable storage medium.
- the existing technology can be Convolutional neural networks analyze CT images to determine tumor size and type.
- the focal loss function is usually used in the commonly used image detection algorithms. This method ignores small tumors and tumors with low contrast, that is, it is not easy to detect small tumors, and there is a problem of inaccurate detection. Therefore, increasing the accuracy of image recognition becomes an urgent problem to be solved.
- a neural network-based image recognition method provided by this application includes:
- the neural network model framework includes an input layer, a hidden layer and an output layer;
- Image recognition is performed on the denoised image data by using the target detection model to obtain a target recognition result.
- the present application also provides an image recognition device based on a neural network, the device comprising:
- the model building module is used to construct a distance-aware loss function by using the pre-built Dice loss function and the Tversky loss function, and use the distance-aware loss function, the pre-built image detection algorithm and the neural network model framework to construct the target object detection to be trained.
- model the neural network model framework includes a data input layer, a hidden layer, and an output layer;
- a model training module configured to obtain a pre-built target image sample set, and use the target image sample set to train the target detection model to be trained to obtain a trained target detection model
- the target object prediction module is used to obtain the CT image to be detected, perform quantization and denoising processing on the CT image to be detected, obtain denoised image data, and use the target detection model to perform the denoising image data on the denoised image data.
- Image recognition to obtain target recognition results.
- the present application also provides an electronic device, the electronic device comprising:
- the memory stores computer program instructions executable by the at least one processor, the computer program instructions being executed by the at least one processor to enable the at least one processor to perform the steps of:
- the neural network model framework includes an input layer, a hidden layer and an output layer;
- Image recognition is performed on the denoised image data by using the target detection model to obtain a target recognition result.
- the present application also provides a computer-readable storage medium, including a storage data area and a storage program area, the storage data area stores created data, and the storage program area stores a computer program; wherein, the computer program is implemented as follows when executed by a processor step:
- the neural network model framework includes an input layer, a hidden layer and an output layer;
- Image recognition is performed on the denoised image data by using the target detection model to obtain a target recognition result.
- FIG. 1 is a schematic flowchart of an image recognition method based on a neural network provided by an embodiment of the present application
- Fig. 2 is a detailed implementation flow chart of a step in the neural network-based image recognition method provided in Fig. 1;
- FIG. 3 is a detailed implementation flow chart of another step in the neural network-based image recognition method provided in FIG. 1;
- FIG. 4 is a schematic block diagram of an image recognition device based on a neural network provided by an embodiment of the present application
- FIG. 5 is a schematic diagram of an internal structure of an electronic device according to an image recognition method based on a neural network provided by an embodiment of the present application.
- the embodiment of the present application provides an image recognition method based on a neural network.
- the execution body of the neural network-based image recognition method includes, but is not limited to, at least one of electronic devices that can be configured to execute the method provided by the embodiments of the present application, such as a server and a terminal.
- the neural network-based image recognition method can be executed by software or hardware installed in a terminal device or a server device, and the software can be a blockchain platform.
- the server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, and the like.
- the neural network-based image recognition method includes:
- the Dice loss function and the Tversky loss function are loss functions for performing regression calculation for the image detection algorithm.
- the Dice loss algorithm can analyze whether each area in the target image contains the features of the target, but often ignore the smaller features, resulting in the omission of the target; the Tversky loss function will cause overtraining in the training process of the model In the case of image detection, overfitting is easy to occur, resulting in false positives such as recognizing other things as targets.
- the target is a tumor.
- the Dice loss function and the Tversky loss function are weighted to construct the distance perception loss function to increase the accuracy of target image recognition.
- the distance-aware loss function constructed by using the pre-built Dice loss function and the Tversky loss function includes:
- Weighted summation is performed on the Dice loss function and the Tversky loss function to obtain the distance-aware loss function, where the distance-aware loss function is:
- L Tversky is the Tversky loss function
- L dis is the Dice loss function
- p k and y k respectively represent the probability value of the target object appearing in each segmented area in the input CT image and the corresponding label of the target object
- ⁇ 1 and ⁇ 2 are the two parameters of the summoning rate ⁇ p k (1-y k ) and the accuracy ⁇ (1-p k )y k respectively
- d k represents each segmented area to the nearest existing target marked as is the distance of the segmented area of the object
- ⁇ represents the number of all pixels in the input CT image
- ⁇ is the weighting coefficient.
- the weighting coefficient ⁇ in the embodiment of the present application can be obtained through the weight analysis between the input layer and the hidden layer in the following training process, and details are not described here.
- the neural network model framework includes an input layer, a hidden layer, and an output layer.
- the target detection model to be trained is constructed by using the distance perception loss function, the pre-built image detection algorithm and the neural network model framework, and the neural network model framework includes an input layer and a hidden layer. and the output layer, including:
- the neural network model framework includes an input layer, a hidden layer, and an output layer;
- the Softmax function is added to the output layer as an output layer activation function to obtain the target object detection model to be trained.
- the neural network model framework is a TransForm model framework, wherein the TransForm model framework includes the input layer, the hidden layer, and the output layer, and the input layer, the hidden layer, and the output layer
- the layers are composed of multiple neurons, and the activation of neurons can perform operations such as data transmission, calculation, and processing.
- the input layer is used for weight configuration of the data, so that when eigenvalues of different orders of magnitude are input to the hidden layer, the eigenvalues of different orders of magnitude can play a role of the same level.
- the hidden layer includes an activation function composed of the image detection algorithm and the distance perception function, and is used for processing the data input by the input layer.
- the output layer classifies the detected objects, and obtains the probability corresponding to each classification result.
- the distance sensing loss function and the image detection algorithm are used to modify the loss function and operation function of the hidden layer
- the Softmax function is used to modify the activation function of the output layer to obtain the target object detection model to be trained.
- the target object image sample set in the embodiment of the present application includes medical images such as CT images and magnetic resonance images of various targets or common tissues, and image analysis tags corresponding to the medical images.
- the target object detection model to be trained is trained by using the target object image sample set, and the trained target object detection model is obtained, including:
- the weight configuration function is an inverse parameter adjustment function, that is, an operation function that changes the numerical input ratio according to the accuracy of the calculation result after training, and retrains until the calculation result is close to the standard answer.
- the activation function is a piecewise function.
- the function When the value of the weight feature value is less than the preset standard value, the function is not activated, and the operation result is 0.
- the function When the value of the weight feature value is greater than or equal to the preset value Standard value, the function activates target detection on the weight feature value.
- the judging whether the minimized loss value converges, and when it is judged that the minimized loss value converges, obtains a final loss value including:
- a minimum loss value will be generated.
- the minimization loss value will gradually decrease with the training process, and the minimization loss value is used to gradually increase the accuracy of the image detection algorithm, and the minimization loss value is The derivation result will gradually decrease.
- the derivation result does not reach the preset value, continue to train the target detection model again by using the target object image sample until the derivation result reaches the The preset value indicates that the downward trend of the minimized loss value disappears, the minimized loss value reaches a convergent state, and the training is completed.
- the target detection model is obtained through the training process from S31 to S34.
- the obtaining of the CT image to be detected, and performing quantization and denoising processing on the CT image to be detected, to obtain denoised image data including:
- the quantized image data is subjected to a Gaussian convolution operation of a two-dimensional neighborhood by using a pre-built Gaussian filter to obtain denoised image data.
- the CT image to be detected is segmented with a preset pixel size, such as 2*2 pixels as the basic unit, to obtain a plurality of pixel blocks, and vectorization processing is performed on the pixel blocks to obtain the quantization process.
- Image data through the pre-built Gaussian filter, perform operations on quantized image data in a square range, such as a 3*3 matrix block, use the obtained operation result to represent the shrinkage value of the 3*3 matrix, and pass the Gaussian filter.
- the convolver in the device operates on adjacent 3*3 matrices until the traversal of the CT image to be detected is completed, and the denoised image data is obtained.
- the image detection is performed on the denoised image data by using the target detection model to obtain a target detection result, including:
- the target in the embodiment of the present application is a tumor
- the denoised image data is imported into the target detection model
- the feature extraction network in the input layer performs feature extraction on the denoised image data
- the extracted Results Match with various tumor cells through the neural network to determine whether there is a tumor in the CT image to be detected.
- texture analysis and category prediction are performed on the characteristics of the tumor through the output layer to accurately determine the tumor type. / or probabilities for individual classes.
- a distance-aware loss function is constructed for the Dice loss function and the Tversky loss function. Since the model trained by the Dice loss function tends to ignore tiny features, the Tversky function tends to be overtrained and misunderstood as a target. Combining the Dice loss function and the Tversky loss function is beneficial to increase the training effect of the model. Therefore, the distance perception loss function is constructed through the Dice loss function and the Tversky loss function, and the perceptual loss function, the image detection algorithm and the neural network model framework are used to construct the target object detection model to be trained, and the target object detection model to be trained is trained.
- the target can be identified more accurately, and the CT image to be detected can be processed by the trained target detection model, and the target can be accurately identified on the CT image to be detected. Therefore, the embodiments of the present application can achieve the purpose of improving the accuracy of target image recognition.
- FIG. 4 it is a schematic block diagram of an image recognition apparatus based on a neural network provided by an embodiment of the present application.
- the image recognition apparatus 100 based on the neural network described in this application can be installed in an electronic device.
- the neural network-based image recognition apparatus may include a model building module 101 , a model training module 102 , and a target object prediction module 103 .
- the modules described in this application may also be referred to as units, which refer to a series of computer program segments that can be executed by the processor of an electronic device and can perform fixed functions, and are stored in the memory of the electronic device.
- each module/unit is as follows:
- the model building module 101 is used to construct a distance-aware loss function using the pre-built Dice loss function and the Tversky loss function, and use the distance-aware loss function, a pre-built image detection algorithm and a neural network model framework to construct a target to be trained Object detection model, the neural network model framework includes an input layer, a hidden layer and an output layer.
- the model building module 101 includes: an algorithm building unit and a model building unit.
- the algorithm construction unit is used for constructing a distance-aware loss function using the pre-built Dice loss function and the Tversky loss function.
- the Dice loss function and the Tversky loss function are loss functions for performing regression calculation for the image detection algorithm.
- the Dice loss algorithm can analyze whether each area in the target image contains the features of the target, but often ignore the smaller features, resulting in the omission of the target; the Tversky loss function will cause overtraining in the training process of the model In the case of image detection, overfitting is easy to occur, resulting in false positives such as recognizing other things as targets.
- the target is a tumor.
- the Dice loss function and the Tversky loss function are weighted to construct the distance perception loss function to increase the accuracy of target image recognition.
- the distance-aware loss function is constructed by using the pre-built Dice loss function and the Tversky loss function, and the algorithm construction unit is used in detail for:
- Weighted summation is performed on the Dice loss function and the Tversky loss function to obtain the distance-aware loss function, where the distance-aware loss function is:
- L Tversky is the Tversky loss function
- L dis is the Dice loss function
- p k and y k respectively represent the probability value of the target object appearing in each segmented area in the input CT image and the corresponding label of the target object
- ⁇ 1 and ⁇ 2 are the two parameters of the summoning rate ⁇ p k (1-y k ) and the accuracy ⁇ (1-p k )y k respectively
- d k represents each segmented area to the nearest existing target marked as is the distance of the segmented area of the object
- ⁇ represents the number of all pixels in the input CT image
- ⁇ is the weighting coefficient.
- the weighting coefficient ⁇ in the embodiment of the present application can be obtained through the weight analysis between the input layer and the hidden layer in the following training process, and details are not described here.
- the model construction unit is used to construct a target object detection model to be trained by using the distance perception loss function, the pre-built image detection algorithm and the neural network model framework, and the neural network model framework includes an input layer, a hidden layer and an output layer .
- the target detection model to be trained is constructed by using the distance perception loss function, the pre-built image detection algorithm and the neural network model framework, and the neural network model framework includes an input layer and a hidden layer. and the output layer, the model building unit is specifically used to:
- the neural network model framework includes an input layer, a hidden layer, and an output layer;
- the Softmax function is added to the output layer as an output layer activation function to obtain the target object detection model to be trained.
- the neural network model framework is a TransForm model framework, wherein the TransForm model framework includes the input layer, the hidden layer, and the output layer, and the input layer, the hidden layer, and the output layer
- the layers are composed of multiple neurons, and the activation of neurons can perform operations such as data transmission, calculation, and processing.
- the input layer is used for weight configuration of the data, so that when eigenvalues of different orders of magnitude are input to the hidden layer, the eigenvalues of different orders of magnitude can play a role of the same level.
- the hidden layer includes an activation function composed of the image detection algorithm and the distance perception function, and is used for processing the data input by the input layer.
- the output layer classifies the detected objects, and obtains the probability corresponding to each classification result.
- the distance sensing loss function and the image detection algorithm are used to modify the loss function and operation function of the hidden layer
- the Softmax function is used to modify the activation function of the output layer to obtain the target object detection model to be trained.
- the model training module 102 is configured to acquire a target object image sample set, and use the target object image sample set to train the target object detection model to be trained to obtain a trained target object detection model.
- the target object image sample set in the embodiment of the present application includes medical images such as CT images and magnetic resonance images of various targets or common tissues, and image analysis tags corresponding to the medical images.
- the target object detection model to be trained is trained by using the target object image sample set, and the trained target object detection model is obtained.
- the model training module 102 Specifically for:
- the target image sample set is imported into the input layer of the target detection model to be trained, and the weight allocation function in the input layer is used to allocate weights to each target image feature in the target image sample set.
- the weight configuration function is an inverse parameter adjustment function, that is, an operation function that changes the numerical input ratio according to the accuracy of the calculation result after training, and retrains until the calculation result is close to the standard answer.
- weight calculation is performed on the tumor image feature to obtain a weight feature value, and the weight feature value is imported into the hidden layer.
- the activation condition of the activation function is triggered by the weight eigenvalue, so that the distance-aware loss function performs a loss operation according to the weight eigenvalue to obtain a minimized loss value, and judges whether the minimized loss value converges. When the minimized loss value converges, the final loss value is obtained.
- the activation function is a piecewise function.
- the function When the value of the weight feature value is less than the preset standard value, the function is not activated, and the operation result is 0.
- the function When the value of the weight feature value is greater than or equal to the preset value Standard value, the function activates target detection on the weight feature value.
- the judging whether the minimized loss value converges, and when it is judged that the minimized loss value converges, obtains a final loss value including:
- a minimized loss value in the training process, each time a new sample is used for training, a minimized loss value will be generated.
- the minimization loss value will gradually decrease with the training process, and the minimization loss value is used to gradually increase the accuracy of the image detection algorithm, and the minimization loss value is The derivation result will gradually decrease.
- the derivation result does not reach the preset value, continue to train the target detection model again by using the target object image sample until the derivation result reaches the The preset value indicates that the downward trend of the minimized loss value disappears, the minimized loss value reaches a convergent state, and the training is completed.
- the image detection algorithm of the activation function in the hidden layer is configured according to the final loss value, the training process is completed, and the trained target detection model is obtained.
- the target object prediction module 103 is used to obtain the CT image to be detected, perform quantization and denoising processing on the CT image to be detected, obtain denoised image data, and use the target object detection model to denoise the noise.
- the image data is used for image recognition, and the target object recognition result is obtained.
- the target object prediction module 103 includes: a detection unit and a judgment unit.
- the detection unit is used for acquiring a CT image to be detected, and performing quantization and denoising processing on the CT image to be detected to obtain denoised image data.
- the acquisition of the CT image to be detected, and the quantization and denoising processing of the CT image to be detected are performed to obtain denoised image data
- the detection unit is specifically used for:
- the quantized image data is subjected to a Gaussian convolution operation of a two-dimensional neighborhood by using a pre-built Gaussian filter to obtain denoised image data.
- a preset pixel size such as 2*2 pixels, is used as the basic unit to segment the CT image to be detected, to obtain a plurality of pixel blocks, and perform vectorization processing on the pixel blocks to obtain the quantization process.
- Image data through the pre-built Gaussian filter, perform operations on the quantized image data in the square range, such as a 3*3 matrix block, use the obtained operation result to represent the shrinkage value of the 3*3 matrix, and pass the Gaussian filter through the Gaussian filter.
- the convolver in the device operates on adjacent 3*3 matrices until the traversal of the CT image to be detected is completed, and the denoised image data is obtained.
- the judging unit is configured to perform image recognition on the denoised image data by using the target detection model to obtain a target recognition result.
- the target detection model is used to detect the denoised image data to obtain a target detection result
- the judgment unit is specifically used for:
- the recognition result is imported into the output layer, so that the output layer activation function determines the probability of the target object and the probability of the corresponding category in the CT image to be detected according to the pre-built classification label of the target object.
- the denoised image data is imported into the target detection model, the feature extraction network in the input layer performs feature extraction on the denoised image data, and the extraction results are combined with various
- the tumor cells are matched to determine whether there is a tumor in the CT image to be detected.
- texture analysis and category prediction are performed on the characteristics of the tumor through the output layer to accurately determine the tumor category/or the probability of each category.
- a distance-aware loss function is constructed for the Dice loss function and the Tversky loss function. Since the model trained by the Dice loss function tends to ignore tiny features, the Tversky function tends to be overtrained and misunderstood as a target. Combining the Dice loss function and the Tversky loss function is beneficial to increase the training effect of the model. Therefore, the distance perception loss function is constructed through the Dice loss function and the Tversky loss function, and the perceptual loss function, the image detection algorithm and the neural network model framework are used to construct the target object detection model to be trained, and the target object detection model to be trained is trained.
- the target can be identified more accurately, and the CT image to be detected can be processed by the trained target detection model, and the target can be accurately identified on the CT image to be detected. Therefore, the embodiments of the present application can achieve the purpose of improving the accuracy of target image recognition.
- FIG. 5 it is a schematic structural diagram of an electronic device for implementing an image recognition method based on a neural network provided by an embodiment of the present application.
- the electronic device 1 may include a processor 10, a memory 11 and a bus, and may also include a computer program stored in the memory 11 and running on the processor 10, such as a neural network-based image recognition method program 12 .
- the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, magnetic disk, CD etc.
- the memory 11 may be an internal storage unit of the electronic device 1 in some embodiments, such as a mobile hard disk of the electronic device 1 .
- the memory 11 may also be an external storage device of the electronic device 1, such as a pluggable mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital) equipped on the electronic device 1. , SD) card, flash memory card (Flash Card), etc.
- the memory 11 may also include both an internal storage unit of the electronic device 1 and an external storage device.
- the memory 11 can not only be used to store application software and various data installed in the electronic device 1, such as the code of the neural network-based image recognition method program 12, etc., but also can be used to temporarily store the data that has been output or will be output. .
- the processor 10 may be composed of integrated circuits, for example, may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits packaged with the same function or different functions, including one or more integrated circuits.
- Central Processing Unit CPU
- microprocessor digital processing chip
- graphics processor and combination of various control chips, etc.
- the processor 10 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect the various components of the entire electronic device, by running or executing the program or module (for example, executing the program) stored in the memory 11.
- Image recognition method program based on neural network, etc.
- the bus may be a peripheral component interconnect (PCI for short) bus or an extended industry standard architecture (Extended industry standard architecture, EISA for short) bus or the like.
- PCI peripheral component interconnect
- EISA Extended industry standard architecture
- the bus can be divided into address bus, data bus, control bus and so on.
- the bus is configured to implement connection communication between the memory 11 and at least one processor 10 and the like.
- FIG. 5 only shows an electronic device with components. Those skilled in the art can understand that the structure shown in FIG. 5 does not constitute a limitation on the electronic device 1, and may include fewer or more components than those shown in the drawings. components, or a combination of certain components, or a different arrangement of components.
- the electronic device 1 may also include a power supply (such as a battery) for powering the various components, preferably, the power supply may be logically connected to the at least one processor 10 through a power management device, so that the power management
- the device implements functions such as charge management, discharge management, and power consumption management.
- the power source may also include one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and any other components.
- the electronic device 1 may further include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.
- the electronic device 1 may also include a network interface, optionally, the network interface may include a wired interface and/or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is usually used in the electronic device 1 Establish a communication connection with other electronic devices.
- a network interface optionally, the network interface may include a wired interface and/or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is usually used in the electronic device 1 Establish a communication connection with other electronic devices.
- the electronic device 1 may further include a user interface, and the user interface may be a display (Display), an input unit (eg, a keyboard (Keyboard)), optionally, the user interface may also be a standard wired interface or a wireless interface.
- the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode, organic light-emitting diode) touch device, and the like.
- the display may also be appropriately called a display screen or a display unit, which is used for displaying information processed in the electronic device 1 and for displaying a visualized user interface.
- the neural network-based image recognition method program 12 stored in the memory 11 in the electronic device 1 is a combination of multiple computer programs, and when running in the processor 10, can realize:
- the neural network model framework includes an input layer, a hidden layer and an output layer;
- Image recognition is performed on the denoised image data by using the target detection model to obtain a target recognition result.
- the modules/units integrated in the electronic device 1 may be stored in a computer-readable storage medium.
- the computer-readable storage medium may be volatile or non-volatile.
- the computer-readable storage medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a removable hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM, Read-Only Memory) Only Memory).
- the computer-readable storage medium may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required by at least one function, and the like; The data created by the use of the node, etc.
- the present application also provides a computer-readable storage medium, where the readable storage medium stores a computer program, and when executed by a processor of an electronic device, the computer program can realize:
- the neural network model framework includes an input layer, a hidden layer and an output layer;
- Image recognition is performed on the denoised image data by using the target detection model to obtain a target recognition result.
- modules described as separate components may or may not be physically separated, and components shown as modules may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment.
- each functional module in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.
- the above-mentioned integrated units can be implemented in the form of hardware, or can be implemented in the form of hardware plus software function modules.
- the blockchain referred to in this application is a new application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
- Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information to verify its Validity of information (anti-counterfeiting) and generation of the next block.
- the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
La présente demande concerne le domaine technique de la détection d'image et divulgue un procédé de reconnaissance d'image basé sur un réseau neuronal, comprenant : la construction d'une fonction de perte de perception de distance à l'aide d'une fonction de perte Dice et d'une fonction de perte Tversky ; l'utilisation de la fonction de perte de perception de distance, d'un algorithme de détection d'image pré-construit, et d'une structure de modèle de réseau neuronal pour construire un modèle de détection d'objet cible à former, la structure de modèle de réseau neuronal comprenant une couche d'entrée, une couche cachée et une couche de sortie ; l'obtention d'un ensemble d'échantillons d'image d'objet cible, et l'utilisation de l'ensemble d'échantillons d'image d'objet cible pour former le modèle de détection d'objet cible à former, pour obtenir un modèle de détection d'objet cible formé ; l'obtention d'une image CT à détecter, et la quantification et le débruitage de ladite image CT pour obtenir des données d'image débruitées ; et la réalisation d'une reconnaissance d'image sur les données d'image débruitées en utilisant le modèle de détection d'objet cible pour obtenir un résultat de reconnaissance d'objet cible. La présente demande concerne en outre un appareil de reconnaissance d'image, un dispositif et un support de stockage lisible par ordinateur. L'objectif de la présente demande est d'améliorer la précision de la reconnaissance d'image d'objet cible.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110378729.2A CN113159147B (zh) | 2021-04-08 | 2021-04-08 | 基于神经网络的图像识别方法、装置、电子设备 |
CN202110378729.2 | 2021-04-08 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022213465A1 true WO2022213465A1 (fr) | 2022-10-13 |
Family
ID=76889200
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2021/097073 WO2022213465A1 (fr) | 2021-04-08 | 2021-05-30 | Procédé et appareil de reconnaissance d'image à base de réseau neuronal, dispositif électronique et support |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN113159147B (fr) |
WO (1) | WO2022213465A1 (fr) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116008911A (zh) * | 2022-12-02 | 2023-04-25 | 南昌工程学院 | 一种基于新型原子匹配准则的正交匹配追踪声源识别方法 |
CN116281479A (zh) * | 2023-04-04 | 2023-06-23 | 南京枫火网络科技有限公司 | 一种基于物联网的电梯故障监测方法及*** |
CN117168863A (zh) * | 2023-10-31 | 2023-12-05 | 苏州市产品质量监督检验院(苏州市质量技术监督综合检验检测中心、苏州市质量认证中心) | 基于人工智能的坐便器防虹吸测试方法、***及存储介质 |
CN117372791A (zh) * | 2023-12-08 | 2024-01-09 | 齐鲁空天信息研究院 | 细粒度定向能毁伤区域检测方法、装置及存储介质 |
CN117671395A (zh) * | 2024-02-02 | 2024-03-08 | 南昌康德莱医疗科技有限公司 | 一种癌细胞种类识别装置 |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113674142B (zh) * | 2021-08-30 | 2023-10-17 | 国家计算机网络与信息安全管理中心 | 图像中目标物的消融方法、装置、计算机设备及介质 |
CN114445410A (zh) * | 2022-04-11 | 2022-05-06 | 武汉飞恩微电子有限公司 | 基于图像识别的电路板检测方法、计算机及可读存储介质 |
CN114676396B (zh) * | 2022-05-30 | 2022-08-30 | 山东极视角科技有限公司 | 深度神经网络模型的保护方法、装置、电子设备和介质 |
CN115439699B (zh) * | 2022-10-25 | 2023-06-30 | 北京鹰瞳科技发展股份有限公司 | 目标检测模型的训练方法、目标检测的方法及相关产品 |
CN116740218B (zh) * | 2023-08-11 | 2023-10-27 | 南京安科医疗科技有限公司 | 一种心脏ct成像图像质量优化方法、设备及介质 |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110232721A (zh) * | 2019-05-16 | 2019-09-13 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | 一种危机器官自动勾画模型的训练方法及装置 |
CN111402268A (zh) * | 2020-03-16 | 2020-07-10 | 苏州科技大学 | 一种医学图像中肝脏及其病灶分割的方法 |
US20210065360A1 (en) * | 2019-08-29 | 2021-03-04 | Varian Medical Systems International Ag | Systems and methods for image cropping and anatomical structure segmentation in medical imaging |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP3432263B1 (fr) * | 2017-07-17 | 2020-09-16 | Siemens Healthcare GmbH | Segmentation sémantique pour la détection du cancer dans la tomosynthèse numérique du sein |
CN110458833B (zh) * | 2019-08-15 | 2023-07-11 | 腾讯科技(深圳)有限公司 | 基于人工智能的医学图像处理方法、医学设备和存储介质 |
ES2813777B2 (es) * | 2019-09-23 | 2023-10-27 | Quibim S L | Metodo y sistema para la segmentacion automatica de hiperintensidades de sustancia blanca en imagenes de resonancia magnetica cerebral |
CN111932482B (zh) * | 2020-09-25 | 2021-05-18 | 平安科技(深圳)有限公司 | 图像中目标物的检测方法、装置、电子设备及存储介质 |
CN112465071A (zh) * | 2020-12-18 | 2021-03-09 | 深圳赛安特技术服务有限公司 | 图像多标签分类方法、装置、电子设备及介质 |
-
2021
- 2021-04-08 CN CN202110378729.2A patent/CN113159147B/zh active Active
- 2021-05-30 WO PCT/CN2021/097073 patent/WO2022213465A1/fr active Application Filing
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110232721A (zh) * | 2019-05-16 | 2019-09-13 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | 一种危机器官自动勾画模型的训练方法及装置 |
US20210065360A1 (en) * | 2019-08-29 | 2021-03-04 | Varian Medical Systems International Ag | Systems and methods for image cropping and anatomical structure segmentation in medical imaging |
CN111402268A (zh) * | 2020-03-16 | 2020-07-10 | 苏州科技大学 | 一种医学图像中肝脏及其病灶分割的方法 |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116008911A (zh) * | 2022-12-02 | 2023-04-25 | 南昌工程学院 | 一种基于新型原子匹配准则的正交匹配追踪声源识别方法 |
CN116008911B (zh) * | 2022-12-02 | 2023-08-22 | 南昌工程学院 | 一种基于新型原子匹配准则的正交匹配追踪声源识别方法 |
CN116281479A (zh) * | 2023-04-04 | 2023-06-23 | 南京枫火网络科技有限公司 | 一种基于物联网的电梯故障监测方法及*** |
CN116281479B (zh) * | 2023-04-04 | 2023-10-17 | 南京枫火网络科技有限公司 | 一种基于物联网的电梯故障监测方法及*** |
CN117168863A (zh) * | 2023-10-31 | 2023-12-05 | 苏州市产品质量监督检验院(苏州市质量技术监督综合检验检测中心、苏州市质量认证中心) | 基于人工智能的坐便器防虹吸测试方法、***及存储介质 |
CN117168863B (zh) * | 2023-10-31 | 2024-01-26 | 苏州市产品质量监督检验院(苏州市质量技术监督综合检验检测中心、苏州市质量认证中心) | 基于人工智能的坐便器防虹吸测试方法、***及存储介质 |
CN117372791A (zh) * | 2023-12-08 | 2024-01-09 | 齐鲁空天信息研究院 | 细粒度定向能毁伤区域检测方法、装置及存储介质 |
CN117372791B (zh) * | 2023-12-08 | 2024-03-22 | 齐鲁空天信息研究院 | 细粒度定向能毁伤区域检测方法、装置及存储介质 |
CN117671395A (zh) * | 2024-02-02 | 2024-03-08 | 南昌康德莱医疗科技有限公司 | 一种癌细胞种类识别装置 |
CN117671395B (zh) * | 2024-02-02 | 2024-04-26 | 南昌康德莱医疗科技有限公司 | 一种癌细胞种类识别装置 |
Also Published As
Publication number | Publication date |
---|---|
CN113159147B (zh) | 2023-09-26 |
CN113159147A (zh) | 2021-07-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2022213465A1 (fr) | Procédé et appareil de reconnaissance d'image à base de réseau neuronal, dispositif électronique et support | |
US11537884B2 (en) | Machine learning model training method and device, and expression image classification method and device | |
US20230385640A1 (en) | Misuse index for explainable artificial intelligence in computing environments | |
WO2022083536A1 (fr) | Procédé et appareil de construction de réseau neuronal | |
US10318848B2 (en) | Methods for object localization and image classification | |
WO2018108129A1 (fr) | Procédé et appareil destinés à l'identification d'un type d'objet, et dispositif électronique | |
WO2018021942A2 (fr) | Reconnaissance faciale au moyen d'un réseau neuronal artificiel | |
WO2022116420A1 (fr) | Procédé et appareil de détection d'événement vocal, dispositif électronique, et support de stockage informatique | |
WO2022105179A1 (fr) | Procédé et appareil de reconnaissance d'image de caractéristiques biologiques, dispositif électronique et support de stockage lisible | |
US11837061B2 (en) | Techniques to provide and process video data of automatic teller machine video streams to perform suspicious activity detection | |
WO2020253508A1 (fr) | Procédé et appareil de détection de cellule anormale, et support d'informations lisible par ordinateur | |
US9750450B2 (en) | Method, electronic apparatus, and computer readable medium of constructing classifier for skin-infection detection | |
WO2021151338A1 (fr) | Procédé d'analyse d'images médicales, appareil, dispositif électronique et support de stockage lisible | |
WO2021189913A1 (fr) | Procédé et appareil de segmentation d'objet cible dans une image, et dispositif électronique et support d'enregistrement | |
US11710552B2 (en) | Method and system for refining label information | |
WO2024001806A1 (fr) | Procédé d'évaluation de données basé sur un apprentissage fédéré et dispositif associé | |
WO2019095587A1 (fr) | Procédé de reconnaissance faciale, serveur d'application et support de stockage lisible par ordinateur | |
CN108492301A (zh) | 一种场景分割方法、终端及存储介质 | |
WO2023231753A1 (fr) | Procédé d'apprentissage de réseau neuronal, procédé de traitement de données et dispositif | |
WO2022127333A1 (fr) | Procédé et appareil d'apprentissage pour un modèle de segmentation d'image, procédé et appareil de segmentation d'image, et dispositif | |
CN110717407A (zh) | 基于唇语密码的人脸识别方法、装置及存储介质 | |
Wu et al. | Pneumonia detection based on RSNA dataset and anchor-free deep learning detector | |
KR20210080762A (ko) | 인공 신경망의 추론 데이터에 대한 신뢰도를 판단하는 방법 | |
WO2022222228A1 (fr) | Procédé et appareil pour reconnaître de mauvaises informations textuelles, et dispositif électronique et support de stockage | |
CN116029760A (zh) | 消息推送方法、装置、计算机设备和存储介质 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 21935686 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 21935686 Country of ref document: EP Kind code of ref document: A1 |