Digital identification method and device based on lightweight network, and medium

Abstract

The invention discloses a digital recognition method and device based on a lightweight network and a medium, and relates to the technical field of intelligent operation and maintenance of rail transit. The method comprises the following steps: constructing a lightweight network, and training based on a nixie tube digital data set to obtain a digital recognition model; and acquiring a real-time image, and inputting each region-of-interest image in the real-time image into the digital recognition model based on a preset region-of-interest of digital recognition to obtain a digital value result. According to the method, the constructed lightweight network is trained by using the historical data set to obtain the digital recognition model, the display data of the nixie tube in the real-time image can be quickly recognized from the real-time image shot by the camera through the preset region of interest based on the digital recognition model, and compared with a traditional manual reading mode, the display data of the nixie tube in the real-time image can be quickly recognized. The accuracy and efficiency of data reading can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of rail transit, and more specifically, to a digital recognition method, device and medium based on a lightweight network. Background Art

[0002] In modern railway systems, pump rooms, as important infrastructure, are responsible for maintaining a stable supply of water and liquids to ensure the normal operation and safety of trains. However, traditional pump room monitoring methods often rely on manual inspections and manual recording, which is not only inefficient but also prone to human errors, affecting the reliability and safety of the system. With the development of digital technology, automated monitoring systems have gradually become an important means to improve the efficiency of pump room management. By using a camera to capture the digital tube numbers in the pump room in real time, automatic monitoring and recording of the equipment status can be achieved. The application of this technology can not only improve the accuracy and real-time nature of data collection, but also reduce manual intervention and management costs.

[0003] However, based on the image data captured by the camera in real time, there is currently a lack of a solution that can quickly identify the digital tube numbers in the image data. Therefore, how to quickly and accurately identify the digital tube numbers in the image is a technical problem that urgently needs to be solved. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a digital recognition method, device and medium based on a lightweight network to quickly and accurately identify the digital tube numbers in the image, thereby improving the accuracy and efficiency of data reading.

[0005] In a first aspect, the present invention provides a digital recognition method based on a lightweight network, the method comprising:

[0006] Build a lightweight network and train it based on the digital tube digital data set to obtain a digital recognition model;

[0007] A real-time image is acquired, and based on a preset digitally recognized region of interest, each region of interest image in the real-time image is input into a digital recognition model to obtain a digital numerical result.

[0008] Furthermore, the lightweight network includes 14 module layers connected in sequence, namely a first module layer, a second module layer, a third module layer, a fourth module layer, a fifth module layer, a sixth module layer, a seventh module layer, an eighth module layer, a ninth module layer, a tenth module layer, an eleventh module layer, a twelfth module layer, a thirteenth module layer and a fourteenth module layer.

[0009] Furthermore, the first module layer is a convolutional layer with a kernel of 3×3, a stride of 2, and an output channel of 32;

[0010] The second module layer is a deep convolutional layer with a kernel of 3×3, a stride of 1, and an output channel of 32;

[0011] The third module layer is a point-by-point convolution layer with a kernel of 1×1, a stride of 1, and an output channel of 64;

[0012] The fourth module layer is a deep convolution layer with a kernel of 3×3, a stride of 1, and an output channel of 64;

[0013] The fifth module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 128;

[0014] The sixth module layer is a deep convolution layer with a kernel of 3×3, a step size of 2, and an output channel of 128;

[0015] The seventh module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 256;

[0016] The eighth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 256;

[0017] The ninth module layer is a point-by-point convolution layer with a kernel of 1×1, a stride of 1, and an output channel of 512;

[0018] The tenth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 512;

[0019] The eleventh module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 1024;

[0020] The twelfth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 1024;

[0021] The thirteenth module layer is an average pooling layer with a kernel of 1×1;

[0022] The fourteenth module layer is a fully connected layer, which after activation is the probability of each predicted label, and the output length is 11.

[0023] Furthermore, training is performed based on the digital tube digital data set to obtain a digital recognition model, including:

[0024] Using the stochastic optimization method of adaptive momentum, the network parameters are continuously updated and the loss function value is recalculated until the loss function value converges or reaches a predetermined number of iterations, to obtain the trained lightweight network, and the trained lightweight network is used as the digital recognition model.

[0025] Further, a real-time image is acquired, and based on a preset digitally recognized region of interest, each region of interest image in the real-time image is input into a digital recognition model to obtain a digital numerical result, including:

[0026] Taking the position where each symbol in the number appears as the region of interest, extracting the region of interest image from the real-time image;

[0027] The region of interest is input into the digital recognition model to obtain a vector of length 11; the first to tenth bits of the vector represent the probability of the numbers 0 to 9, and the eleventh bit represents the probability of the decimal point; the label corresponding to the highest probability is a string of a single number in the current region of interest;

[0028] The single digital detection results of each region of interest image are sequentially spliced ​​together to obtain a character string of digital detection results as a digital value result.

[0029] In a second aspect, the present invention provides a digital recognition device based on a lightweight network, the device comprising:

[0030] The model building module is configured to build a lightweight network, train it based on the digital tube digital data set, and obtain a digital recognition model;

[0031] The image recognition module is configured to acquire a real-time image, and based on a preset digitally recognized region of interest, input each region of interest image in the real-time image into a digital recognition model to obtain a digital numerical result.

[0032] Furthermore, the lightweight network includes 14 module layers connected in sequence, namely a first module layer, a second module layer, a third module layer, a fourth module layer, a fifth module layer, a sixth module layer, a seventh module layer, an eighth module layer, a ninth module layer, a tenth module layer, an eleventh module layer, a twelfth module layer, a thirteenth module layer and a fourteenth module layer.

[0033] Furthermore, the first module layer is a convolutional layer with a kernel of 3×3, a stride of 2, and an output channel of 32;

[0034] The second module layer is a deep convolutional layer with a kernel of 3×3, a stride of 1, and an output channel of 32;

[0035] The third module layer is a point-by-point convolution layer with a kernel of 1×1, a stride of 1, and an output channel of 64;

[0036] The fourth module layer is a deep convolution layer with a kernel of 3×3, a stride of 1, and an output channel of 64;

[0037] The fifth module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 128;

[0038] The sixth module layer is a deep convolution layer with a kernel of 3×3, a step size of 2, and an output channel of 128;

[0039] The seventh module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 256;

[0040] The eighth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 256;

[0041] The ninth module layer is a point-by-point convolution layer with a kernel of 1×1, a stride of 1, and an output channel of 512;

[0042] The tenth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 512;

[0043] The eleventh module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 1024;

[0044] The twelfth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 1024;

[0045] The thirteenth module layer is an average pooling layer with a kernel of 1×1;

[0046] The fourteenth module layer is a fully connected layer, which after activation is the probability of each predicted label, and the output length is 11.

[0047] Furthermore, the image recognition module is further configured to:

[0048] Taking the position where each symbol in the number appears as the region of interest, extracting the region of interest image from the real-time image;

[0049] The region of interest is input into the digital recognition model to obtain a vector of length 11; the first to tenth bits of the vector represent the probability of the numbers 0 to 9, and the eleventh bit represents the probability of the decimal point; the label corresponding to the highest probability is a string of a single number in the current region of interest;

[0050] The single digital detection results of each region of interest image are sequentially spliced ​​together to obtain a character string of digital detection results as a digital value result.

[0051] In a third aspect, the present invention provides a readable storage medium, wherein the readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the method as described above.

[0052] The present invention has at least the following beneficial effects:

[0053] The present invention uses historical data sets to train the constructed lightweight network to obtain a digital recognition model. Based on the digital recognition model, the display data of the digital tube in the real-time image captured by the camera can be quickly identified through a preset region of interest. Compared with the traditional manual reading method, the accuracy and efficiency of data reading can be effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A flowchart of a digital recognition method based on a lightweight network according to an embodiment of the present invention is shown.

[0055] Figure 2 A flowchart of digital value recognition according to an embodiment of the present invention is shown.

[0056] Figure 3 The output result of the digital recognition model according to the embodiment of the present invention is shown.

[0057] Figure 4 A structural diagram of a digital recognition device based on a lightweight network according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0058] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and specific embodiments, but are not intended to limit the present invention. For the various steps described herein, if there is no necessity for a causal relationship between each other, the order in which they are described as examples herein should not be regarded as a limitation, and those skilled in the art should know that they can be adjusted in order, as long as the logic between them is not destroyed, resulting in the inability to implement the entire process.

[0059] The embodiment of the present invention provides a digital recognition method based on a lightweight network. Figure 1 The figure is a flow chart of a digital recognition method based on a lightweight network. The digital recognition method based on a lightweight network includes steps S10 to S20, which are described in detail as follows.

[0060] S10: Build a lightweight network and train it based on the digital tube digital dataset to obtain a digital recognition model.

[0061] In this embodiment, the lightweight network includes 14 module layers connected in sequence, namely the first module layer, the second module layer, the third module layer, the fourth module layer, the fifth module layer, the sixth module layer, the seventh module layer, the eighth module layer, the ninth module layer, the tenth module layer, the eleventh module layer, the twelfth module layer, the thirteenth module layer and the fourteenth module layer.

[0062] Specifically, the first module layer is a convolution layer with a kernel of 3×3, a step size of 2, and an output channel of 32; the second module layer is a depth convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 32; the third module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 64; the fourth module layer is a depth convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 64; the fifth module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 128; the sixth module layer is a depth convolution layer with a kernel of 3×3, a step size of 2, and an output channel of 128; the seventh module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 2 56; the eighth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 256; the ninth module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 512; the tenth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 512; the eleventh module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 1024; the twelfth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 1024; the thirteenth module layer is an average pooling layer with a kernel of 1×1; the fourteenth module layer is a fully connected layer, which is the probability of each predicted label after activation, and the output length is 11.

[0063] In some embodiments, training is performed based on a digital tube digital data set to obtain a digital recognition model, including: using a random optimization method with adaptive momentum, continuously updating network parameters, recalculating the loss function value, until the loss function value converges or reaches a predetermined number of iterations, obtaining a trained lightweight network, and using the trained lightweight network as a digital recognition model.

[0064] For example, a track data set is selected as a digital tube digital data set, which consists of pictures and labels, each picture has only a single number or decimal point, and this embodiment has 12,000 pictures and labels, with a training set ratio of 80% and a test set ratio of 20%. During training, the Adam optimizer is used for network training, the predetermined number of iterations is 800 steps, and the learning rate is 0.0002.

[0065] S20: acquiring a real-time image, and based on a preset digitally recognized region of interest, inputting each region of interest image in the real-time image into a digital recognition model to obtain a digital numerical result.

[0066] In some embodiments, Figure 2 As shown in FIG. 1 , it is a flowchart of digital value recognition, and step S20 is specifically implemented by the following steps:

[0067] S21: taking the position where each symbol in the number appears as the region of interest, extracting the region of interest image from the real-time image;

[0068] S22: Input the region of interest into the digital recognition model to obtain a vector with a length of 11; the first to tenth bits of the vector represent the probability of the numbers 0 to 9, and the eleventh bit represents the probability of the decimal point; the label corresponding to the highest probability is a string of a single number in the current region of interest;

[0069] S23: sequentially concatenate the single digital detection results of the images of the respective regions of interest to obtain a character string of digital detection results as a digital value result.

[0070] For example, an example result is as follows Figure 3 As shown, the image size is 300*500, and 4 regions of interest are set, such as Figure 3 As shown in the box, the model outputs a total of 5 labels, namely "0", ".", "3", "4", and "8" respectively. After splicing, the detection result is 0.348.

[0071] The embodiment of the present invention also provides a digital recognition device based on a lightweight network, such as Figure 4 As shown, the device comprises:

[0072] The model building module 401 is configured to build a lightweight network, perform training based on a digital tube digital data set, and obtain a digital recognition model;

[0073] The image recognition module 402 is configured to acquire a real-time image, and based on a preset digitally recognized region of interest, input each region of interest image in the real-time image into a digital recognition model to obtain a digital numerical result.

[0074] In some embodiments, the lightweight network includes 14 module layers connected in sequence, namely a first module layer, a second module layer, a third module layer, a fourth module layer, a fifth module layer, a sixth module layer, a seventh module layer, an eighth module layer, a ninth module layer, a tenth module layer, an eleventh module layer, a twelfth module layer, a thirteenth module layer and a fourteenth module layer.

[0075] In some embodiments, the first module layer is a convolutional layer with a kernel of 3×3, a stride of 2, and an output channel of 32;

[0076] The second module layer is a deep convolutional layer with a kernel of 3×3, a stride of 1, and an output channel of 32;

[0077] The third module layer is a point-by-point convolution layer with a kernel of 1×1, a stride of 1, and an output channel of 64;

[0078] The fourth module layer is a deep convolution layer with a kernel of 3×3, a stride of 1, and an output channel of 64;

[0079] The fifth module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 128;

[0080] The sixth module layer is a deep convolution layer with a kernel of 3×3, a step size of 2, and an output channel of 128;

[0081] The seventh module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 256;

[0082] The eighth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 256;

[0083] The ninth module layer is a point-by-point convolution layer with a kernel of 1×1, a stride of 1, and an output channel of 512;

[0084] The tenth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 512;

[0085] The eleventh module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 1024;

[0086] The twelfth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 1024;

[0087] The thirteenth module layer is an average pooling layer with a kernel of 1×1;

[0088] The fourteenth module layer is a fully connected layer, which after activation is the probability of each predicted label, and the output length is 11.

[0089] In some embodiments, the image recognition module is further configured to:

[0090] Taking the position where each symbol in the number appears as the region of interest, extracting the region of interest image from the real-time image;

[0091] The region of interest is input into the digital recognition model to obtain a vector of length 11; the first to tenth bits of the vector represent the probability of the numbers 0 to 9, and the eleventh bit represents the probability of the decimal point; the label corresponding to the highest probability is a string of a single number in the current region of interest;

[0092] The single digital detection results of each region of interest image are sequentially spliced ​​together to obtain a character string of digital detection results as a digital value result.

[0093] It should be noted that the structures of the various lightweight network-based digital recognition devices described in this embodiment belong to the same technical concept as the previously described lightweight network-based digital recognition method, and achieve the same beneficial effects through the same principles, which will not be repeated here.

[0094] An embodiment of the present invention further provides a readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the method described in any of the above embodiments.

[0095] In addition, although exemplary embodiments have been described herein, the scope includes any and all embodiments based on the present invention with equivalent elements, modifications, omissions, combinations (e.g., various embodiments intersecting schemes), adaptations or changes. The elements in the claims will be interpreted broadly based on the language adopted in the claims, and are not limited to the examples described in this specification or during the implementation of this application, and the examples will be interpreted as non-exclusive. Therefore, this specification and examples are intended to be considered as examples only, and the true scope and spirit are indicated by the following claims and the full scope of their equivalents.

[0096] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. For example, those of ordinary skill in the art can use other embodiments when reading the above description. In addition, in the above-mentioned specific embodiments, various features can be grouped together to simplify the present invention. This should not be interpreted as a feature of an invention that is not claimed for protection being necessary for any claim. On the contrary, the subject matter of the present invention may be less than all the features of the embodiments of a specific invention. Thus, the following claims are incorporated into the specific embodiments as examples or embodiments, wherein each claim is independently used as a separate embodiment, and it is considered that these embodiments can be combined with each other in various combinations or arrangements. The scope of the present invention should be determined with reference to the full scope of the equivalent forms of the attached claims and these claims.

Claims

1. A digital recognition method based on a lightweight network, characterized in that: The method comprises: Build a lightweight network and train it based on the digital tube digital data set to obtain a digital recognition model; A real-time image is acquired, and based on a preset digitally recognized region of interest, each region of interest image in the real-time image is input into a digital recognition model to obtain a digital numerical result.

2. The digital recognition method based on lightweight network according to claim 1 is characterized in that: The lightweight network includes 14 module layers connected in sequence, namely a first module layer, a second module layer, a third module layer, a fourth module layer, a fifth module layer, a sixth module layer, a seventh module layer, an eighth module layer, a ninth module layer, a tenth module layer, an eleventh module layer, a twelfth module layer, a thirteenth module layer and a fourteenth module layer.

3. The digital recognition method based on lightweight network according to claim 2 is characterized in that: The first module layer is a convolutional layer with a kernel of 3×3, a stride of 2, and an output channel of 32; The second module layer is a deep convolutional layer with a kernel of 3×3, a stride of 1, and an output channel of 32; The third module layer is a point-by-point convolution layer with a kernel of 1×1, a stride of 1, and an output channel of 64; The fourth module layer is a deep convolution layer with a kernel of 3×3, a stride of 1, and an output channel of 64; The fifth module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 128; The sixth module layer is a deep convolution layer with a kernel of 3×3, a step size of 2, and an output channel of 128; The seventh module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 256; The eighth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 256; The ninth module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 512; The tenth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 512; The eleventh module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 1024; The twelfth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 1024; The thirteenth module layer is an average pooling layer with a kernel of 1×1; The fourteenth module layer is a fully connected layer, which after activation is the probability of each predicted label, and the output length is 11.

4. The digital recognition method based on lightweight network according to claim 2 is characterized in that: Based on the digital tube digital data set, the digital recognition model is trained, including: Using the stochastic optimization method of adaptive momentum, the network parameters are continuously updated and the loss function value is recalculated until the loss function value converges or reaches a predetermined number of iterations, to obtain the trained lightweight network, and the trained lightweight network is used as the digital recognition model.

5. The digital recognition method based on lightweight network according to claim 4 is characterized in that: Acquire a real-time image, and based on a preset digitally recognized region of interest, input each region of interest image in the real-time image into a digital recognition model to obtain a digital numerical result, including: Taking the position where each symbol in the number appears as the region of interest, extracting the region of interest image from the real-time image; The region of interest is input into the digital recognition model to obtain a vector of length 11; the first to tenth bits of the vector represent the probability of the numbers 0 to 9, and the eleventh bit represents the probability of the decimal point; the label corresponding to the highest probability is a string of a single number in the current region of interest; The single digital detection results of each region of interest image are sequentially spliced ​​together to obtain a character string of digital detection results as a digital value result.

6. A digital recognition device based on a lightweight network, characterized in that: The device comprises: The model building module is configured to build a lightweight network, train it based on the digital tube digital data set, and obtain a digital recognition model; The image recognition module is configured to acquire a real-time image, and based on a preset digitally recognized region of interest, input each region of interest image in the real-time image into a digital recognition model to obtain a digital numerical result.

7. The digital recognition device based on a lightweight network according to claim 6, characterized in that: The lightweight network includes 14 module layers connected in sequence, namely a first module layer, a second module layer, a third module layer, a fourth module layer, a fifth module layer, a sixth module layer, a seventh module layer, an eighth module layer, a ninth module layer, a tenth module layer, an eleventh module layer, a twelfth module layer, a thirteenth module layer and a fourteenth module layer.

8. The digital recognition device based on lightweight network according to claim 7, characterized in that: The first module layer is a convolutional layer with a kernel of 3×3, a stride of 2, and an output channel of 32; The second module layer is a deep convolutional layer with a kernel of 3×3, a stride of 1, and an output channel of 32; The third module layer is a point-by-point convolution layer with a kernel of 1×1, a stride of 1, and an output channel of 64; The fourth module layer is a deep convolution layer with a kernel of 3×3, a stride of 1, and an output channel of 64; The fifth module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 128; The sixth module layer is a deep convolution layer with a kernel of 3×3, a step size of 2, and an output channel of 128; The seventh module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 256; The eighth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 256; The ninth module layer is a point-by-point convolution layer with a kernel of 1×1, a stride of 1, and an output channel of 512; The tenth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 512; The eleventh module layer is a point-by-point convolution layer with a kernel of 1×1, a step size of 1, and an output channel of 1024; The twelfth module layer is a deep convolution layer with a kernel of 3×3, a step size of 1, and an output channel of 1024; The thirteenth module layer is an average pooling layer with a kernel of 1×1; The fourteenth module layer is a fully connected layer, which after activation is the probability of each predicted label, and the output length is 11.

9. The digital recognition device based on a lightweight network according to claim 6, characterized in that: The image recognition module is further configured to: Taking the position where each symbol in the number appears as the region of interest, extracting the region of interest image from the real-time image; The region of interest is input into the digital recognition model to obtain a vector of length 11; the first to tenth bits of the vector represent the probability of the numbers 0 to 9, and the eleventh bit represents the probability of the decimal point; the label corresponding to the highest probability is a string of a single number in the current region of interest; The single digital detection results of each region of interest image are sequentially spliced ​​together to obtain a character string of digital detection results as a digital value result. 10 . A non-transitory computer-readable storage medium storing instructions, which, when executed by a processor, executes the method according to claim 1 .

Images