Motor state identification method, device, equipment and storage medium
Through the combination of deep learning recognition model and pre-trained feature extraction network, the problems of high cost and low efficiency in motor fault detection are solved, and the efficient identification of motor status is achieved, which is suitable for complex industrial environments.
Patent Information
- Application Number
- CN202111223604.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-10-20
- Publication Date
- 2024-12-10
- Estimated Expiration
- 2041-10-20
AI Technical Summary
Existing motor fault detection methods require the acquisition of high-precision motor images, which are cost-effective and inefficient, and cannot be applied to complex industrial environments.
The deep learning recognition model is used to combine pre-trained feature extraction networks to obtain motor image features through segmentation networks and classification network training to achieve efficient recognition of motor states.
It reduces the detection cost of motor status recognition, improves detection efficiency, and is suitable for complex industrial environments.
Smart Images
Figure CN114022419B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of equipment detection, and in particular to a motor state recognition method, device, equipment and storage medium. Background Art
[0002] In industrial environments, motors are used more and more frequently and are increasingly closely associated with human production activities. As an indispensable power provider in people's production and life, many motors will experience abnormal failures during use, which will have a lot of impact on production.
[0003] Currently, most motor fault detection methods require obtaining high-precision motor images, which have high detection costs and low detection efficiency, and cannot be applied to today's increasingly complex usage environments. Summary of the invention
[0004] The present invention provides a motor state recognition method, device, equipment and storage medium to solve the defects of the prior art that high-precision motor images need to be obtained, the detection cost is high, and the detection efficiency is low, so as to realize efficient recognition of the motor state.
[0005] In a first aspect, the present invention provides a method for identifying a motor state, comprising:
[0006] Acquire an image of a motor to be identified;
[0007] Inputting the motor image to be identified into a pre-trained feature extraction network to obtain image features of the motor image to be identified;
[0008] The image features of the motor image to be identified are input into a deep learning recognition model to obtain a motor state recognition result of the motor image to be identified.
[0009] According to a motor state recognition method provided by the present invention, the deep learning recognition model includes a segmentation network and a classification network, and the deep learning recognition model is trained and acquired through the following steps:
[0010] Acquire image feature training samples and motor states corresponding to the image feature training samples;
[0011] Inputting the image feature training sample into the segmentation network to obtain a first branch feature map and a second branch feature map;
[0012] Stacking the first branch feature map and the second branch feature map to obtain a first feature map;
[0013] Inputting the first feature map into the classification network to obtain a sample motor state recognition result of the image feature training sample;
[0014] Training verification is performed according to the motor state corresponding to the image feature training sample and the sample motor state recognition result of the image feature training sample to obtain the deep learning recognition model.
[0015] A motor state identification method provided by the present invention also includes:
[0016] The loss function of the deep learning recognition model is determined according to the loss function of the segmentation network and / or the loss function of the classification network.
[0017] According to a motor state recognition method provided by the present invention, the loss function of the deep learning recognition model is:
[0018] L=α·β·L1+(1-α)·γ·L2
[0019] α=1-n / n_epoch
[0020] Wherein, L represents the loss function of the deep learning recognition model, L1 represents the loss function of the segmentation network, and L2 represents the loss function of the classification network;
[0021] Wherein, β represents the first adjustment parameter, when the input image feature training sample contains samples with pixel-level labels, β is 1; when the input image feature training sample does not contain samples with pixel-level labels, β is 0;
[0022] Wherein, γ represents the second adjustment parameter, which is used to adjust the ratio of L1 and L2 in L;
[0023] Among them, n represents the number of training times, and n_epoch represents the total number of training cycles.
[0024] According to a motor state recognition method provided by the present invention, the image feature training sample is input into the segmentation network to obtain a first branch feature map and a second branch feature map, including:
[0025] The image feature training sample undergoes one or more convolution operations to obtain the first branch feature map;
[0026] The image feature training samples are subjected to one or more maximum pooling operations to obtain the second branch feature map.
[0027] According to a motor state recognition method provided by the present invention, the first feature map is input into the classification network to obtain a sample motor state recognition result of the image feature training sample, including:
[0028] The first feature map is subjected to three convolution operations and three maximum pooling operations to obtain a second feature map;
[0029] The first feature map and the second feature map are respectively subjected to setting processing to obtain recognition feature maps, wherein the setting processing includes an average pooling operation and / or a maximum pooling operation;
[0030] The recognition feature map is input into a fully connected layer to obtain a sample motor state recognition result of the image feature training sample.
[0031] In a second aspect, the present invention further provides a motor state identification device, comprising:
[0032] An image acquisition unit, used for acquiring an image of a motor to be identified;
[0033] A feature extraction unit, used for inputting the motor image to be identified into a pre-trained feature extraction network to obtain image features of the motor image to be identified;
[0034] The recognition result unit is used to input the image features of the motor image to be recognized into the deep learning recognition model to obtain the motor state recognition result of the motor image to be recognized.
[0035] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the motor state identification method as described in any one of the first aspects above are implemented.
[0036] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the motor state identification method as described in any one of the first aspects above.
[0037] In a fifth aspect, the present invention further provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of the motor state identification method as described in any one of the first aspects above are implemented.
[0038] The motor state recognition method, device, equipment and storage medium provided by the present invention obtain image features of the motor image to be identified through a pre-trained feature extraction network, input the image features of the motor image to be identified into a deep learning recognition model, obtain the motor state recognition result of the motor image to be identified, and realize efficient recognition of the motor state. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0040] Figure 1 It is a flow chart of the motor state identification method provided by the present invention;
[0041] Figure 2 It is a structural schematic diagram of the motor state identification device provided by the present invention;
[0042] Figure 3 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION
[0043] In the present invention, the term "and / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.
[0044] In the present invention, the term "plurality" refers to two or more than two, and other quantifiers are similar to them.
[0045] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0046] Combine the following Figure 1 The motor state identification method of the present invention is described.
[0047] Figure 1 is a flow chart of the motor state identification method provided by the present invention, such as Figure 1 As shown, the motor state identification method provided by the present invention may include:
[0048] Step 101: Acquire an image of a motor to be identified.
[0049] Specifically, in an environment of industrial production of motors, an industrial camera may be used to collect images of motors in an actual production process to obtain images of motors to be identified.
[0050] When the motor is running, the motor image can be photographed by a high-speed electronic scanner to obtain the motor image to be identified.
[0051] The format of the motor image is not specifically limited, and may be a Tag Image File Format (TIFF), a Graphics Interchange Format (GIF), or a Joint Photographic Expert Group (JPEG).
[0052] Step 102: input the motor image to be identified into a pre-trained feature extraction network to obtain image features of the motor image to be identified.
[0053] Specifically, the pre-trained feature extraction network can determine the features of the input object according to the input object. When the input object is a motor image, the pre-trained feature extraction network can output motor image features corresponding to the motor image according to the input motor image.
[0054] The motor image to be identified is input into a pre-trained feature extraction network, and the pre-trained feature extraction network outputs image features of the motor image to be identified.
[0055] The pre-trained feature extraction network can be pre-trained based on a fully convolutional network (FCN), can be pre-trained based on a spatial regularization (SRN) network, or can be pre-trained based on a spatial transformer network (STN).
[0056] Step 103: input the image features of the motor image to be identified into the deep learning recognition model to obtain the click state recognition result of the motor image to be identified.
[0057] Specifically, the deep learning recognition model is used to determine the motor state recognition result corresponding to the input image feature according to the input image feature. According to the motor state, the motor state can be divided into three categories: major abnormality, other abnormality and compliance with the standard. The motor state recognition result can also be divided into three categories: major abnormality, other abnormality and compliance with the standard.
[0058] The image features of the motor image to be identified are input into the deep learning recognition model, and the output result of the deep learning recognition model is the motor state recognition result of the motor image to be identified.
[0059] It can be seen from the above embodiments that the motor state recognition method provided by the present invention obtains the image features of the motor image to be identified through a pre-trained feature extraction network, inputs the image features into a deep learning recognition model, obtains the motor state recognition result of the motor image to be identified, and realizes efficient recognition of the motor state.
[0060] Optionally, the deep learning recognition model includes a segmentation network and a classification network, and the deep learning recognition model is trained and acquired through the following steps:
[0061] Acquire image feature training samples and motor states corresponding to the image feature training samples;
[0062] Inputting the image feature training sample into the segmentation network to obtain a first branch feature map and a second branch feature map;
[0063] Stacking the first branch feature map and the second branch feature map to obtain a first feature map;
[0064] Inputting the first feature map into the classification network to obtain a sample motor state recognition result of the image feature training sample;
[0065] Training verification is performed according to the motor state corresponding to the image feature training sample and the sample motor state recognition result of the image feature training sample to obtain the deep learning recognition model.
[0066] Specifically, the deep learning recognition model can include a segmentation network and a classification network.
[0067] When training a deep learning recognition model, it is necessary to obtain image feature training samples and motor states corresponding to the image feature training samples. The image feature training samples may include training samples where most motor states meet the standards, training samples where a small number of motor states are other abnormalities, and training samples where a small number of motor states are major abnormalities. The motor state corresponding to each training sample in the image feature training samples is predetermined to facilitate training verification with the output results of the deep learning recognition model.
[0068] The image feature training samples are input into the segmentation network, and the segmentation network outputs the first branch feature map and the second branch feature map to obtain the first branch feature map and the second branch feature map. The segmentation network performs additional supervision on the image feature training samples, which enables the deep learning recognition model to better obtain the feature information of the image feature map.
[0069] The first branch feature map and the second branch feature map are stacked to obtain a first feature map.
[0070] The first feature map is input into the classification network, and the classification network outputs the sample motor state recognition result of the image feature training sample to obtain the sample motor state recognition result of the image feature training sample.
[0071] The image feature training samples are obtained through the deep learning recognition model to obtain the sample motor state recognition results of the image feature training samples, and the motor states corresponding to the image feature training samples and the sample motor state recognition results of the image feature training samples are compared, analyzed, verified and trained, and the parameters of the deep learning recognition model are further adjusted to obtain the deep learning recognition model with good performance in the final verification result, and the deep learning recognition model with good performance is selected as the deep learning recognition model, without specific limitation here.
[0072] It can be seen from the above embodiments that the motor state recognition method provided by the present invention can further supervise the image features of the motor image to be identified through the segmentation network of the deep learning recognition model, thereby increasing the accuracy of motor state recognition.
[0073] Optionally, the motor state identification method further includes:
[0074] The loss function of the deep learning recognition model is determined according to the loss function of the segmentation network and / or the loss function of the classification network.
[0075] Specifically, the loss function of the segmentation network is predetermined and may be a mean square error loss function; the loss function of the classification network is also predetermined and may be a cross entropy loss function. The loss function of the deep learning recognition model may be determined by the loss function of the segmentation network and the loss function of the classification network, or may be determined only by the loss function of the segmentation network, or may be determined only by the loss function of the classification network.
[0076] It can be seen from the above embodiments that the motor state recognition method provided by the present invention determines the loss function of the deep learning recognition model by the loss function of the segmentation network and / or the loss function of the classification network, thereby further improving the image features of the motor image to be identified and increasing the accuracy of motor state recognition.
[0077] Optionally, the loss function of the deep learning recognition model is:
[0078] L=α·β·L1+(1-α)·γ·L2
[0079] α=1-n / n_epoch
[0080] Wherein, L represents the loss function of the deep learning recognition model, L1 represents the loss function of the segmentation network, and L2 represents the loss function of the classification network;
[0081] Wherein, β represents the first adjustment parameter, when the input image feature training sample contains samples with pixel-level labels, β is 1; when the input image feature training sample does not contain samples with pixel-level labels, β is 0;
[0082] Wherein, γ represents the second adjustment parameter, which is used to adjust the ratio of L1 and L2 in L;
[0083] Among them, n represents the number of training times, and n_epoch represents the total number of training cycles.
[0084] Specifically, the loss function of the deep learning recognition model is:
[0085] L=α·β·L1+(1-α)·γ·L2
[0086] α=1-n / n_epoch
[0087] Among them, L represents the loss function of the deep learning recognition model, L1 represents the loss function of the segmentation network, and L2 represents the loss function of the classification network.
[0088] Among them, β is an adjustment parameter. When the input image is a sample without abnormalities or a sample containing pixel-level labels, β is 1, that is, the entire deep learning recognition model focuses on both the segmentation network and the classification network. When the input image does not contain pixel-level label samples, that is, the input image only contains label categories, β is 0, and the entire network only focuses on the classification network.
[0089] Wherein, γ represents the second adjustment parameter, and the adjustment range is between 0 and 1. By changing the value of γ, the ratio of L1 and L2 in L is adjusted. In the training process of the image feature map training sample, specific presets can be made, and generally the value of γ is 1.
[0090] Among them, n represents the number of training times, and n_epoch represents the total number of training cycles. For example, when the number of image feature map training samples is 1000, when training to the 100th sample, n represents 100, n_epoch represents 1000, and α=1-n / n_epoch means that when training the image feature map training samples, the loss function of the entire deep learning recognition model in the first stage focuses more on the segmentation network, and the loss function of the entire deep learning recognition model in the second stage focuses more on the classification network.
[0091] It can be seen from the above embodiments that the motor state recognition method provided by the present invention is determined by the specific functional relationship between the loss function of the segmentation network and the loss function of the classification network through the loss function of the deep learning recognition model, thereby further improving the image features of the motor image to be identified and increasing the accuracy of motor state recognition.
[0092] Optionally, inputting the image feature training sample into the segmentation network to obtain a first branch feature map and a second branch feature map includes:
[0093] The image feature map training sample undergoes one or more convolution operations to obtain the first branch feature map;
[0094] The image feature map training sample is subjected to one or more maximum pooling operations to obtain the second branch feature map;
[0095] Specifically, the image feature map training samples are input into the segmentation network. In the segmentation network, the operations that can be used are Conv2D convolution operation and Max-pool maximum pooling operation. The size of the convolution kernel of the convolution layer is 5x5 and 1x1, and the size of the convolution kernel of the maximum pooling layer is 2x2. The input image feature map training samples can be processed through five convolution operations and three maximum pooling operations. The convolution operation is used to extract features, and the pooling operation is used to reduce the amount of parameters to avoid overfitting.
[0096] The image feature map training sample undergoes one or more convolution operations to obtain a first branch feature map.
[0097] The image feature map training samples undergo one or more maximum pooling operations to obtain the second branch feature map.
[0098] For example, two convolution operations are performed with a convolution kernel size of 5x5, and the input image is converted from 3 channels to 32 channels to obtain the first branch feature map. After downsampling through a maximum pooling layer, the number of channels remains unchanged at 32, and the second branch feature map is obtained.
[0099] Three convolution operations were performed with a convolution kernel size of 5x5 and the number of channels changed from 32 to 64 to obtain the first branch feature map. After downsampling through a maximum pooling layer, the number of channels remained unchanged at 64 to obtain the second branch feature map.
[0100] It can be seen from the above embodiments that the motor state recognition method provided by the present invention further obtains the image features of the motor image to be identified by inputting the image feature training samples into the segmentation network, and undergoes one or more convolution operations and maximum pooling operations in the segmentation network, thereby increasing the accuracy of motor state recognition.
[0101] Optionally, inputting the first feature map into the classification network to obtain a sample motor state recognition result of the image feature training sample includes:
[0102] The first feature map is subjected to three convolution operations and three maximum pooling operations to obtain a second feature map;
[0103] The first feature map and the second feature map are respectively subjected to setting processing to obtain recognition feature maps, wherein the setting processing includes an average pooling operation and / or a maximum pooling operation;
[0104] The recognition feature map is input into a fully connected layer to obtain a sample motor state recognition result of the image feature training sample.
[0105] Specifically, the first feature map is input into the classification network, and undergoes three convolution operations and three maximum pooling operations in the classification network to obtain the second feature map.
[0106] The first feature map and the second feature map are set and processed. The setting process may include an average pooling operation and a maximum pooling operation, or may only include an average pooling operation, or may only include a maximum pooling operation.
[0107] The recognition feature map is input into the fully connected layer of the classification network for recognition, and the sample motor state recognition result of the image feature training sample is obtained.
[0108] For example, the operations used in the classification network are Conv2D convolution operation and Max-pool maximum pooling operation, where the size of the convolution kernel of the convolution layer is 5x5, and the size of the convolution kernel of the maximum pooling layer is 2x2.
[0109] The first feature map is input into the classification network and undergoes three convolution operations and three maximum pooling operations to obtain the second feature map.
[0110] The first time, downsampling is performed through a maximum pooling operation, and the number of channels remains unchanged at 32. Then a convolution operation is performed with a convolution kernel size of 5x5, and the input image is converted from 1025 channels to 8 channels.
[0111] The second time, downsampling is performed through a maximum pooling operation, the number of channels remains unchanged at 8, and then a convolution operation is performed with a convolution kernel size of 5x5, and the input image is converted from 8 channels to 16 channels.
[0112] The third time, downsampling is performed through a maximum pooling operation, the number of channels remains unchanged at 16, and then a convolution operation is performed with a convolution kernel size of 5x5. The input image is converted from 16 channels to 32 channels, and the output is the second feature map.
[0113] Finally, the second feature map is subjected to an Average pooling layer downsampling operation, and the output is Yaver. A Max pooling layer downsampling operation is performed, and the output is Ymax. The number of channels of Yaver and Ymax is 32. For the first feature map, an Average pooling layer downsampling operation is performed, and the output is Xaver. A Max pooling layer downsampling operation is performed, and the output is Xmax. The number of channels of Xaver and Xmax is 1. Then the four outputs Xaver, Xmax, Yaver, and Ymax are stacked to form an output with a channel number of 66. The output is then passed through a Fully connected layer to form an output with a channel number of 1, and the sample motor state recognition result of the image feature training sample is obtained.
[0114] It can be seen from the above embodiments that the motor state recognition method provided by the present invention obtains a second feature map by performing three convolution operations and three maximum pooling operations on the first feature map in the classification network, further obtains the image features of the motor image to be identified, and increases the accuracy of motor state recognition.
[0115] The motor state identification device provided by the present invention is described below. The motor state identification device described below and the motor state identification method described above can be referred to each other.
[0116] Figure 2 is a schematic diagram of the structure of the motor state identification device provided by the present invention, such as Figure 2 As shown, the motor state identification device of the present invention includes:
[0117] An image acquisition unit 201 is used to acquire an image of a motor to be identified;
[0118] A feature extraction unit 202, used for inputting the motor image to be identified into a pre-trained feature extraction network to obtain image features of the motor image to be identified;
[0119] The recognition result unit 203 is used to input the image features of the motor image to be recognized into the deep learning recognition model to obtain the motor state recognition result of the motor image to be recognized.
[0120] Figure 3 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 3As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330 and a communication bus 340, wherein the processor 310, the communication interface 320 and the memory 330 communicate with each other through the communication bus 340. The processor 310 may call the logic instructions in the memory 330 to execute the motor state identification method, which includes:
[0121] Acquire an image of a motor to be identified;
[0122] Inputting the motor image to be identified into a pre-trained feature extraction network to obtain image features of the motor image to be identified;
[0123] The image features of the motor image to be identified are input into a deep learning recognition model to obtain a motor state recognition result of the motor image to be identified.
[0124] In addition, the logic instructions in the above-mentioned memory 330 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.
[0125] On the other hand, the present invention further provides a computer program product, the computer program product includes a computer program, the computer program can be stored in a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the motor state identification method provided by the above methods, the method includes:
[0126] Acquire an image of a motor to be identified;
[0127] Inputting the motor image to be identified into a pre-trained feature extraction network to obtain image features of the motor image to be identified;
[0128] The image features of the motor image to be identified are input into a deep learning recognition model to obtain a click state recognition result of the motor image to be identified.
[0129] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the motor state identification method provided by the above methods is implemented, and the method includes:
[0130] Acquire an image of a motor to be identified;
[0131] Inputting the motor image to be identified into a pre-trained feature extraction network to obtain image features of the motor image to be identified;
[0132] The image features of the motor image to be identified are input into a deep learning recognition model to obtain a click state recognition result of the motor image to be identified.
[0133] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.
[0134] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for identifying a motor state, characterized in that: include: Acquire an image of a motor to be identified; Inputting the motor image to be identified into a pre-trained feature extraction network to obtain image features of the motor image to be identified; Inputting the image features of the motor image to be identified into a deep learning recognition model to obtain a motor state recognition result of the motor image to be identified; Wherein, the deep learning recognition model includes a segmentation network and a classification network, and the deep learning recognition model is trained and acquired through the following steps: acquiring an image feature training sample and a motor state corresponding to the image feature training sample; inputting the image feature training sample into the segmentation network to obtain a first branch feature map and a second branch feature map; stacking the first branch feature map and the second branch feature map to obtain a first feature map; inputting the first feature map into the classification network to obtain a sample motor state recognition result of the image feature training sample; performing training verification according to the motor state corresponding to the image feature training sample and the sample motor state recognition result of the image feature training sample to obtain the deep learning recognition model; Wherein, the loss function of the deep learning recognition model is determined according to the loss function of the segmentation network and / or the loss function of the classification network; Among them, the loss function of the deep learning recognition model is: L=α·β·L1+(1-α)·γ·L2 α=1-n / n_epoch Wherein, L represents the loss function of the deep learning recognition model, L1 represents the loss function of the segmentation network, and L2 represents the loss function of the classification network; wherein, β represents the first adjustment parameter, when the input image feature training sample contains samples with pixel-level labels, β is 1; when the input image feature training sample does not contain samples with pixel-level labels, β is 0; wherein, γ represents the second adjustment parameter, which is used to adjust the ratio of L1 and L2 in L; wherein n represents the number of training times, and n_epoch represents the total number of training cycles.
2. The motor state identification method according to claim 1, characterized in that: The step of inputting the image feature training sample into the segmentation network to obtain a first branch feature map and a second branch feature map comprises: The image feature training sample undergoes one or more convolution operations to obtain the first branch feature map; The image feature training samples are subjected to one or more maximum pooling operations to obtain the second branch feature map.
3. The motor state identification method according to claim 1, characterized in that: The step of inputting the first feature map into the classification network to obtain a sample motor state recognition result of the image feature training sample includes: The first feature map is subjected to three convolution operations and three maximum pooling operations to obtain a second feature map; The first feature map and the second feature map are respectively subjected to setting processing to obtain recognition feature maps, wherein the setting processing includes an average pooling operation and / or a maximum pooling operation; The recognition feature map is input into a fully connected layer to obtain a sample motor state recognition result of the image feature training sample.
4. A motor state identification device, characterized in that: include: An image acquisition unit, used for acquiring an image of a motor to be identified; A feature extraction unit, used for inputting the motor image to be identified into a pre-trained feature extraction network to obtain image features of the motor image to be identified; A recognition result unit, used for inputting the image features of the motor image to be recognized into a deep learning recognition model to obtain a motor state recognition result of the motor image to be recognized; Wherein, the deep learning recognition model includes a segmentation network and a classification network, and the recognition result unit is further used to obtain an image feature training sample and a motor state corresponding to the image feature training sample; the image feature training sample is input into the segmentation network to obtain a first branch feature map and a second branch feature map; the first branch feature map and the second branch feature map are stacked to obtain a first feature map; the first feature map is input into the classification network to obtain a sample motor state recognition result of the image feature training sample; training verification is performed according to the motor state corresponding to the image feature training sample and the sample motor state recognition result of the image feature training sample to obtain the deep learning recognition model; Wherein, the recognition result unit is further used to determine the loss function of the deep learning recognition model according to the loss function of the segmentation network and / or the loss function of the classification network; Among them, the loss function of the deep learning recognition model is: L=α·β·L1+(1-α)·γ·L2 α=1-n / n_epoch Wherein, L represents the loss function of the deep learning recognition model, L1 represents the loss function of the segmentation network, and L2 represents the loss function of the classification network; wherein, β represents the first adjustment parameter, when the input image feature training sample contains samples with pixel-level labels, β is 1; when the input image feature training sample does not contain samples with pixel-level labels, β is 0; wherein, γ represents the second adjustment parameter, which is used to adjust the ratio of L1 and L2 in L; wherein n represents the number of training times, and n_epoch represents the total number of training cycles.
5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the motor state identification method according to any one of claims 1 to 3 are implemented.
6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the motor state identification method according to any one of claims 1 to 3 are implemented.
7. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the motor state identification method according to any one of claims 1 to 3 are implemented.