A neural network training method for detecting the optical purification effect of organic waste gas

Through deep neural network and computer vision technology, the optical purification effect of organic exhaust gas is detected, and the differential map and position mask matrix screening characteristics are used to solve the problems of cumbersome detection process and poor real-time performance in the existing technology, and high-accurate optical Purification effect monitoring.

CN112541462BActive Publication Date: 2024-12-06HANGZHOU ZHUILIE TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202011519366.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-12-21
Publication Date
2024-12-06
Estimated Expiration
2040-12-21

AI Technical Summary

Technical Problem

The prior art is difficult to detect the photocatalytic oxidation effect in industrially emitted organic waste gases in real time and accurately, especially due to changes in optical characteristics and the complexity of particulate impurities, the detection process is cumbersome and real-time cannot be guaranteed.

Method used

Using computer vision technology based on deep neural networks, optical images are extracted through convolutional neural networks, and feature screening through differential maps and position mask matrix, focusing on image feature expressions other than particles, improving the accuracy of light purification effect detection Rate.

Benefits of technology

It realizes rapid and accurate detection of the light purification effect of organic waste gas, improves the accuracy and real-timeness of detection, and simplifies the detection process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112541462B_ABST
    Figure CN112541462B_ABST
Patent Text Reader

Abstract

The present application discloses a training method for a neural network for detecting the light purification effect of organic waste gas. In the training method, an image that does not contain gas impurities but contains particulate matter impurities is used as a reference image, and is input into two convolutional neural networks with the same structure as the training image, thereby obtaining a training feature map and a reference feature map, and then determining the difference between the feature values ​​of each position in the training feature map and the corresponding position in the reference feature map, and determining each position where the difference is less than a predetermined threshold from the training feature map, and then deleting the feature values ​​of these positions, thereby obtaining a feature map that can focus on the expression of other image features other than particulate matter.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to a neural network training method for detecting the light purification effect of organic waste gas, a method for detecting the light purification effect of organic waste gas based on a deep neural network, a neural network training system for detecting the light purification effect of organic waste gas, a light purification effect detection system and electronic equipment for organic waste gas based on a deep neural network. Background Art

[0002] There are many types of organic waste gas emitted by industry, with complex composition, irregular concentration changes and huge flow rate. The main treatment method at present is to treat it through photocatalytic oxidation combined with particulate matter filtration technology, that is, by using the light emitted by UV light source to irradiate the waste gas and contact the waste gas efficiently, thereby removing organic gas impurities in the waste gas, and then filtering out particulate matter in the gas through filtering materials.

[0003] Relatively speaking, the filtering effect of particulate matter is relatively easy to detect, while the photocatalytic oxidation effect is currently mainly achieved by sampling the gas and performing chemical testing. This requires complex experimental operations on the organic gases therein, and the process is cumbersome and cannot guarantee real-time performance.

[0004] Therefore, an optimized solution for detecting the optical purification effect of organic waste gas is desired.

[0005] At present, deep learning and neural networks have been widely used in computer vision, natural language processing, text signal processing and other fields. In addition, deep learning and neural networks have also shown a level close to or even beyond that of humans in image classification, object detection, semantic segmentation, text translation and other fields.

[0006] In recent years, the development of deep learning and neural networks has provided new solutions and solutions for the detection of optical purification effects of organic waste gas. Summary of the invention

[0007] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a neural network training method for detecting the light purification effect of organic waste gas, a method for detecting the light purification effect of organic waste gas based on a deep neural network, a neural network training system for detecting the light purification effect of organic waste gas, a light purification effect detection system for organic waste gas based on a deep neural network, and an electronic device, which uses computer vision technology based on artificial intelligence to detect the photocatalytic oxidation effect based on optical images, and when using a convolutional neural network to extract features from an image, the obtained feature map is subjected to feature screening processing, so that the processed feature map can focus more on other image feature expressions other than particulate matter, and in this way, the accuracy of light purification effect detection is improved.

[0008] According to one aspect of the present application, a neural network training method for detecting the light purification effect of organic waste gas is provided, which comprises:

[0009] Acquire a plurality of first images of the organic waste gas after photocatalytic oxidation under the irradiation of a UV light source and a plurality of second images of the organic waste gas after photocatalytic oxidation and detected to contain no organic gas but particulate impurities under the irradiation of the same UV light source;

[0010] Passing the plurality of first images through a first convolutional neural network to obtain a training feature map;

[0011] Passing the plurality of second images through a second convolutional neural network to obtain a reference feature map, wherein the first convolutional neural network and the second convolutional neural network have the same network structure;

[0012] Calculating the pixel-wise difference between the training feature map and the reference feature map to obtain a difference map;

[0013] Determine a position set where the value of each pixel position in the difference image is less than a preset threshold to obtain a position mask matrix;

[0014] Based on the position mask matrix, deleting the value of the corresponding position in the training feature map to obtain an enhanced feature map;

[0015] Passing the enhanced feature map through a classifier to obtain a classification loss function value; and

[0016] Based on the classification loss function value, the parameters of the first convolutional neural network and the classifier are updated.

[0017] In the above-mentioned training method of a neural network for detecting the light purification effect of organic waste gas, a set of positions in which the value of each pixel position in the differential map is less than a preset threshold is determined to obtain a position mask matrix, including: in response to the value of the pixel position in the differential map being less than the preset threshold, the corresponding position is set to 0; in response to the value of the pixel position in the differential map being greater than or equal to the preset threshold, the corresponding position is set to 1; and the encoding values ​​of the corresponding pixel positions are arranged in order to obtain the position mask matrix.

[0018] In the above-mentioned neural network training method for detecting the light purification effect of organic waste gas, the preset threshold is involved in the training process of the first convolutional neural network as a hyperparameter.

[0019] In the above-mentioned neural network training method for detecting the light purification effect of organic waste gas, the values ​​of corresponding positions in the training feature map are deleted based on the position mask matrix to obtain an enhanced feature map, including: multiplying the position mask matrix by the training feature map at pixel position to obtain the enhanced feature map.

[0020] In the above-mentioned training method of a neural network for detecting the light purification effect of organic waste gas, the enhanced feature map is passed through a classifier to obtain a classification loss function value, including: passing the enhanced feature map through multiple convolutional layers to obtain a deep feature map; passing the deep feature map through one or more fully connected layers to obtain a classification feature vector; and passing the classification feature vector through a Softmax classification function to obtain the classification loss function value.

[0021] In the above-mentioned neural network training method for detecting the light purification effect of organic waste gas, the first convolutional neural network and the second convolutional neural network are deep residual networks.

[0022] According to another aspect of the present application, a method for detecting the light purification effect of organic waste gas based on a deep neural network is provided, which includes:

[0023] Acquire an image to be detected, where the image to be detected is an image collected after the organic waste gas is photocatalytically oxidized under irradiation of a UV light source; and

[0024] The image to be detected is input into the first convolutional neural network and classifier trained according to the neural network training method for detecting the light purification effect of organic waste gas as described above to obtain a classification result, which is used to indicate whether the light purification effect of the organic waste gas is qualified.

[0025] According to another aspect of the present application, a neural network training system for detecting the optical purification effect of organic waste gas is provided, which comprises:

[0026] A training image acquisition unit, used to acquire a plurality of first images of the organic waste gas after photocatalytic oxidation under the irradiation of a UV light source and a plurality of second images of the organic waste gas after photocatalytic oxidation and detected to contain no organic gas but particulate impurities under the irradiation of the same UV light source;

[0027] a training feature map generating unit, configured to pass the plurality of first images obtained by the training image obtaining unit through a first convolutional neural network to obtain a training feature map;

[0028] a reference feature map generating unit, configured to pass the plurality of second images obtained by the training image obtaining unit through a second convolutional neural network to obtain a reference feature map, wherein the first convolutional neural network and the second convolutional neural network have the same network structure;

[0029] a difference map generating unit, configured to calculate a pixel-wise difference between the training feature map obtained by the training feature map generating unit and the reference feature map obtained by the reference feature map generating unit, so as to obtain a difference map;

[0030] A position mask matrix generating unit, used to determine a position set where the value of each pixel position in the difference image obtained by the difference image generating unit is less than a preset threshold, so as to obtain a position mask matrix;

[0031] an enhanced feature map generating unit, configured to delete the value of the corresponding position in the training feature map based on the position mask matrix obtained by the position mask matrix generating unit to obtain an enhanced feature map;

[0032] a classification loss function value calculation unit, configured to pass the enhanced feature map obtained by the enhanced feature map generation unit through a classifier to obtain a classification loss function value; and

[0033] A parameter updating unit is used to update the parameters of the first convolutional neural network and the classifier based on the classification loss function value obtained by the classification loss function value calculation unit.

[0034] In the above-mentioned neural network training system for detecting the light purification effect of organic waste gas, the position mask matrix generation unit further includes: in response to the value of the pixel position in the differential map being less than a preset threshold, setting the corresponding position to 0; in response to the value of the pixel position in the differential map being greater than or equal to the preset threshold, setting the corresponding position to 1; and arranging the encoding values ​​of the corresponding pixel positions in order to obtain the position mask matrix.

[0035] In the above-mentioned neural network training system for detecting the light purification effect of organic waste gas, the preset threshold is involved in the training process of the first convolutional neural network as a hyperparameter.

[0036] In the above-mentioned neural network training system for detecting the light purification effect of organic waste gas, the enhanced feature map generation unit is further used to: perform point multiplication of the position mask matrix and the training feature map at pixel positions to obtain the enhanced feature map.

[0037] In the above-mentioned neural network training system for detecting the light purification effect of organic waste gas, the classification loss function value calculation unit includes: a convolution processing subunit, which is used to pass the enhanced feature map through multiple convolution layers to obtain a deep feature map; a fully connected processing subunit, which is used to pass the deep feature map through one or more fully connected layers to obtain a classification feature vector; and a loss function calculation subunit, which is used to pass the classification feature vector through a Softmax classification function to obtain the classification loss function value.

[0038] In the above-mentioned neural network training system for detecting the light purification effect of organic waste gas, the first convolutional neural network and the second convolutional neural network are deep residual networks.

[0039] According to another aspect of the present application, a light purification effect detection system for organic waste gas based on a deep neural network is also provided, which includes:

[0040] An image acquisition unit to be detected, used to acquire an image to be detected, wherein the image to be detected is an image collected under UV light source after the organic waste gas is photocatalytically oxidized; and

[0041] A classification result generating unit is used to input the image to be detected obtained by the image acquisition unit to the first convolutional neural network and classifier trained according to the training method of the neural network for detecting the optical purification effect of organic waste gas as described above, so as to obtain a classification result, wherein the classification result is used to indicate whether the optical purification effect of the organic waste gas is qualified.

[0042] According to another aspect of the present application, there is provided an electronic device, comprising: a processor; and a memory, in which computer program instructions are stored, and when the computer program instructions are executed by the processor, the processor executes the training method of a neural network for monitoring the capacity of bagged cement in a cement packaging machine as described above, or a method for detecting the optical purification effect of organic waste gas based on a deep neural network.

[0043] According to another aspect of the present application, a computer-readable medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the processor executes the training method of a neural network for monitoring the capacity of bagged cement in a cement packaging machine as described above, or a method for detecting the optical purification effect of organic waste gas based on a deep neural network.

[0044] According to the neural network training method for detecting the light purification effect of organic waste gas, the light purification effect detection method for organic waste gas based on deep neural network, the neural network training system for detecting the light purification effect of organic waste gas, the light purification effect detection system and electronic device for organic waste gas based on deep neural network provided in the present application, the computer vision technology based on artificial intelligence is used to detect the photocatalytic oxidation effect based on optical images, and when using convolutional neural network to extract features from images, the obtained feature map is subjected to feature screening processing so that the processed feature map can focus more on other image feature expressions other than particulate matter. In this way, the accuracy of light purification effect detection is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0046] Figure 1 The diagram illustrates a schematic diagram of an application scenario of a neural network training method for detecting the optical purification effect of organic waste gas and a method for detecting the optical purification effect of organic waste gas based on a deep neural network according to an embodiment of the present application.

[0047] Figure 2 A flowchart of a neural network training method for detecting the optical purification effect of organic waste gas according to an embodiment of the present application is illustrated.

[0048] Figure 3 The figure shows a schematic diagram of the architecture of a neural network training method for detecting the optical purification effect of organic waste gas according to an embodiment of the present application.

[0049] Figure 4 The diagram illustrates a flow chart of deleting the values ​​of corresponding positions in the training feature map based on the position mask matrix to obtain an enhanced feature map in a training method of a neural network for detecting the light purification effect of organic waste gas according to an embodiment of the present application.

[0050] Figure 5 The diagram illustrates a flow chart of a neural network training method for detecting the light purification effect of organic waste gas according to an embodiment of the present application, in which the enhanced feature map is passed through a classifier to obtain a classification loss function value.

[0051] Figure 6The figure illustrates a flow chart of a method for detecting the optical purification effect of organic waste gas based on a deep neural network according to an embodiment of the present application.

[0052] Figure 7 The diagram illustrates a block diagram of a neural network training system for detecting the optical purification effect of organic waste gas according to an embodiment of the present application.

[0053] Figure 8 The diagram illustrates a block diagram of a classification loss function value calculation unit in a neural network training system for detecting the optical purification effect of organic waste gas according to an embodiment of the present application.

[0054] Fig. 9 The figure illustrates a block diagram of a light purification effect detection system for organic waste gas based on a deep neural network according to an embodiment of the present application.

[0055] Fig.10 A block diagram of an electronic device according to an embodiment of the present application is illustrated. DETAILED DESCRIPTION

[0056] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0057] Scenario Overview

[0058] As mentioned above, industrial organic waste gas is of many types, with complex composition, irregular concentration changes, and huge flow rate. The current main treatment method is to treat it through photocatalytic oxidation combined with particulate matter filtration technology, that is, by using the light emitted by a UV light source to irradiate the waste gas and efficiently contact the waste gas, thereby removing organic gas impurities in the waste gas, and then filtering out particulate matter in the gas through filtering materials.

[0059] Relatively speaking, the filtering effect of particulate matter is relatively easy to detect, while the photocatalytic oxidation effect is currently mainly achieved by sampling the gas and performing chemical testing. This requires complex experimental operations on the organic gases therein, and the process is cumbersome and cannot guarantee real-time performance.

[0060] Therefore, an optimized solution for detecting the optical purification effect of organic waste gas is desired.

[0061] The applicant of the present application has discovered that when removing organic gas impurities in exhaust gas through photocatalytic oxidation, due to changes in the chemical composition of the air, the optical properties of the gas will also change accordingly under the same UV light source irradiation. Therefore, it is possible to consider using artificial intelligence-based computer vision technology to detect the photocatalytic oxidation effect based on optical images.

[0062] In addition, due to the presence of particulate impurities in the image, when using convolutional neural networks to extract features from the image, it is hoped that the extracted features can focus on the expression of other image features other than the particles. Therefore, it is necessary to perform feature screening on the obtained feature map.

[0063] This feature screening process can be achieved by a process similar to the difference operation of the set, that is, an image that does not contain gas impurities but contains particulate impurities is used as a reference image, and is input into two convolutional neural networks with the same structure as the training image, thereby obtaining a training feature map and a reference feature map, and then determining the difference between the feature values ​​of each position in the training feature map and the corresponding position in the reference feature map, and determining each position where the difference is less than a predetermined threshold from the training feature map, and then deleting the feature values ​​of these positions. Here, the predetermined threshold is to allow the convolutional neural network to retain a predetermined tolerance during feature extraction. In addition, the corresponding positions after deletion can be filled with any identical default values, thereby obtaining a feature map that can focus on the expression of other image features other than particulate matter. In this way, after the feature map is further passed through multiple convolutional layers and classifiers, the accuracy of the detection of the photocatalytic oxidation effect can be improved.

[0064] Based on this, the present application proposes a neural network training method for detecting the light purification effect of organic waste gas, which includes: obtaining multiple first images of organic waste gas after photocatalytic oxidation under UV light source irradiation and multiple second images of organic waste gas after photocatalytic oxidation and detection that does not contain organic gas but contains particulate impurities under the same UV light source irradiation; passing the multiple first images through a first convolutional neural network to obtain a training feature map; passing the multiple second images through a second convolutional neural network to obtain a reference feature map, and the first convolutional neural network and the second convolutional neural network have the same network structure; calculating the pixel position difference between the training feature map and the reference feature map to obtain a difference map; determining a set of positions where the value of each pixel position in the difference map is less than a preset threshold to obtain a position mask matrix; based on the position mask matrix, deleting the value of the corresponding position in the training feature map to obtain an enhanced feature map; passing the enhanced feature map through a classifier to obtain a classification loss function value; and, based on the classification loss function value, updating the parameters of the first convolutional neural network and the classifier.

[0065] Based on this, the present application also proposes a method for detecting the optical purification effect of organic waste gas based on a deep neural network, which includes: obtaining an image to be detected, wherein the image to be detected is an image collected under the irradiation of a UV light source after the organic waste gas is photocatalytically oxidized; and inputting the image to be detected into a first convolutional neural network and a classifier trained according to the training method of a neural network for detecting the optical purification effect of organic waste gas as described above, to obtain a classification result, wherein the classification result is used to indicate whether the optical purification effect of the organic waste gas is qualified.

[0066] Figure 1 The diagram illustrates a schematic diagram of an application scenario of a neural network training method for detecting the optical purification effect of organic waste gas and a method for detecting the optical purification effect of organic waste gas based on a deep neural network according to an embodiment of the present application.

[0067] like Figure 1 As shown, in the training phase of this application scenario, a camera (e.g., Figure 1 C) collecting a plurality of first images of the organic waste gas after photocatalytic oxidation under the irradiation of a UV light source and a plurality of second images of the organic waste gas after photocatalytic oxidation and detected to contain no organic gas but particulate impurities under the irradiation of the same UV light source; then, the plurality of first images and the plurality of second images are input into a server in which a training algorithm of a neural network for detecting the light purification effect of organic waste gas is deployed (for example, as Figure 2 S) as shown in , wherein the server is capable of training the neural network for light purification effect detection using the multiple first images and the multiple second images using a training algorithm of a neural network for light purification effect detection of organic waste gas.

[0068] After training is completed, in the detection phase, firstly, the camera (e.g. Figure 1 C) as shown in the figure, the image to be detected is acquired, and the image to be detected is an image collected under UV light source after the organic waste gas is photocatalytically oxidized; then, the image to be detected is input into a server deployed with a light purification effect detection algorithm for organic waste gas based on a deep neural network (for example, Figure 2 S) as shown in the figure, wherein the server can process the acquired image to be detected using a light purification effect detection algorithm for organic waste gas based on a deep neural network to generate a classification result indicating whether the light purification effect of the organic waste gas is qualified.

[0069] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0070] Exemplary Methods

[0071] Figure 2 The flowchart of the neural network training method for detecting the light purification effect of organic waste gas according to the embodiment of the present application is shown. Figure 2 As shown, according to an embodiment of the present application, a neural network training method for detecting the light purification effect of organic waste gas includes: S110, obtaining a plurality of first images of organic waste gas after photocatalytic oxidation under the irradiation of a UV light source and a plurality of second images of organic waste gas after photocatalytic oxidation and detected to contain no organic gas but particulate impurities under the irradiation of the same UV light source; S120, passing the plurality of first images through a first convolutional neural network to obtain a training feature map, S130, passing the plurality of second images through a second convolutional neural network to obtain a reference feature map, the first convolutional neural network and the second convolutional neural network are respectively used to obtain a training feature map. The convolution neural network has the same network structure; S140, calculating the pixel position difference between the training feature map and the reference feature map to obtain a difference map; S150, determining a set of positions where the value of each pixel position in the difference map is less than a preset threshold to obtain a position mask matrix; S160, based on the position mask matrix, deleting the values ​​of corresponding positions in the training feature map to obtain an enhanced feature map; S170, passing the enhanced feature map through a classifier to obtain a classification loss function value; and, S180, based on the classification loss function value, updating the parameters of the first convolution neural network and the classifier.

[0072] Figure 3 The figure shows a schematic diagram of the architecture of a neural network training method for detecting the light purification effect of organic waste gas according to an embodiment of the present application. Figure 3 As shown, in the network architecture, firstly, a plurality of first images (for example, Figure 3 IN0) is passed through a first convolutional neural network (e.g., Figure 3 ) to obtain the training feature map (e.g., Figure 3 At the same time, a plurality of second images (for example, as shown in F1) of the organic waste gas obtained after photocatalytic oxidation and detected to contain no organic gas but particulate impurities under the same UV light source are taken. Figure 3 IN1) is passed through a second convolutional neural network (e.g., Figure 3 ) to obtain a reference feature map (e.g., Figure 3 The first convolutional neural network and the second convolutional neural network have the same network structure. Next, the pixel position difference between the training feature map and the reference feature map is calculated to obtain a difference map (for example, Figure 3Then, a position set where the value of each pixel position in the difference image is less than a preset threshold is determined to obtain a position mask matrix (for example, as Figure 3 Mp as shown in FIG), and based on the position mask matrix, the value of the corresponding position in the training feature map is deleted to obtain an enhanced feature map (for example, Figure 3 Then, the enhanced feature map is passed through a classifier to obtain a classification loss function value. Then, based on the classification loss function value, the parameters of the first convolutional neural network and the classifier are updated.

[0073] In step S110, a plurality of first images of the organic waste gas after photocatalytic oxidation under the irradiation of a UV light source and a plurality of second images of the organic waste gas after photocatalytic oxidation and detected to contain no organic gas but particulate impurities under the irradiation of the same UV light source are obtained. As mentioned above, in the technical solution of the present application, the inventor of the present application found that when removing organic gas impurities in the waste gas by photocatalytic oxidation, due to the change in the chemical composition of the air, under the irradiation of the same UV light source, the optical properties presented by the gas will also change accordingly. Therefore, it is possible to consider using computer vision technology based on artificial intelligence to detect the photocatalytic oxidation effect based on optical images. Computer vision technology based on artificial intelligence can be used to detect the photocatalytic oxidation effect based on optical images.

[0074] In addition, due to the presence of particulate matter impurities in the optical image, when using the convolutional neural network to extract features from the image, it is more desirable that the extracted features can focus on the expression of other image features other than particulate matter, so it is necessary to perform feature screening on the obtained feature map. This feature screening process can be achieved by a process similar to the difference operation of a set, so in the technical solution of the present application, a second image that does not contain gas impurities but contains particulate matter impurities is used as a reference image.

[0075] In step S120, the plurality of first images are passed through a first convolutional neural network to obtain a training feature map. That is, a deep convolutional neural network is used as a feature extractor to extract high-dimensional features in the first images. Here, the first images contain both organic gas and particulate impurities, so the training feature map includes both high-dimensional feature expressions of organic gas and high-dimensional feature expressions of particulate impurities.

[0076] Those of ordinary skill in the art should know that deep convolutional neural networks have excellent performance in extracting local spatial features of images. In a specific example of the present application, the first convolutional neural network is implemented as a deep residual network, for example, ResNet 50. Those of ordinary skill in the art should know that, compared with traditional convolutional neural networks, deep residual networks are an optimized network structure proposed on the basis of traditional convolutional neural networks, which mainly solves the problem of gradient disappearance during training. The deep residual network introduces a residual network structure, through which the network layer can be made deeper, and the problem of gradient disappearance will not occur. The residual network draws on the cross-layer link idea of ​​the high-speed network, which breaks the convention that the traditional neural network can only give N layers as input from the input layer of the N-1 layer, so that the output of a certain layer can directly cross several layers as the input of a certain layer behind it. Its significance is to provide a new direction for the problem that the error rate of the entire learning model increases instead of decreasing by superimposing multiple layers of networks.

[0077] In step S130, the plurality of second images are passed through a second convolutional neural network to obtain a reference feature map, and the first convolutional neural network and the second convolutional neural network have the same network structure. That is, the second convolutional neural network having the same network structure is used to extract high-dimensional features in the second image. Here, it should be understood that because the second image does not contain organic gas but contains particulate impurities, the reference feature map has a high-dimensional feature expression of particulate impurities but does not have a high-dimensional feature expression of organic gas.

[0078] In step S140, the pixel position difference between the training feature map and the reference feature map is calculated to obtain a difference map. By calculating the pixel position difference between the training feature map and the reference feature map, the same feature expressions of the training feature map and the reference feature map in the high-dimensional feature space can be eliminated to focus on the difference features between the training feature map and the reference feature map to obtain the difference feature map.

[0079] Specifically, as mentioned above, the reference feature map has a high-dimensional feature expression of particulate matter impurities but does not have a high-dimensional feature expression of organic gases. The training feature map includes both high-dimensional feature expressions of organic gases and high-dimensional feature expressions of particulate matter impurities. Therefore, the difference map can focus more on the high-dimensional feature expression of organic gases and reduce the high-dimensional feature expression of particulate matter impurities.

[0080] In step S150, a position set in which the value of each pixel position in the difference map is less than a preset threshold is determined to obtain a position mask matrix. It should be understood that the position set in the difference map that is less than the preset threshold is more likely to be a particle feature expression in the semantic space, and therefore, the position mask matrix represents the position of the high-dimensional feature distribution of the particle in the training feature map.

[0081] More specifically, in an embodiment of the present application, the process of determining a set of positions in which the value of each pixel position in the differential image is less than a preset threshold to obtain a position mask matrix includes: first, in response to the value of the pixel position in the differential image being less than the preset threshold, setting the corresponding position to 0; in response to the value of the pixel position in the differential image being greater than or equal to the preset threshold, setting the corresponding position to 1; and, arranging the encoding values ​​of the corresponding pixel positions in order to obtain the position mask matrix.

[0082] In step S160, based on the position mask matrix, the values ​​of the corresponding positions in the training feature map are deleted to obtain an enhanced feature map. That is, based on the position mask matrix, each position where the difference is less than a predetermined threshold is determined from the training feature map, and then the eigenvalues ​​of these positions are deleted. Specifically, the position where the eigenvalue is 0 in the position mask matrix is ​​the position to be deleted.

[0083] Specifically, in an embodiment of the present application, the process of deleting the value of the corresponding position in the training feature map based on the position mask matrix to obtain the enhanced feature map includes: performing a dot multiplication of the position mask matrix and the training feature map by pixel position to obtain the enhanced feature map. Moreover, the corresponding position after deletion can be filled with any same default value, so as to obtain a feature map that can focus on the expression of other image features other than particulate matter.

[0084] Figure 4 The figure shows a flow chart of deleting the values ​​of corresponding positions in the training feature map based on the position mask matrix to obtain an enhanced feature map in a neural network training method for detecting the light purification effect of organic waste gas according to an embodiment of the present application. Figure 4 As shown, in an embodiment of the present application, the values ​​of corresponding positions in the training feature map are deleted based on the position mask matrix to obtain an enhanced feature map, including: S210, multiplying the position mask matrix by the training feature map at pixel positions to obtain the enhanced feature map; and, S220, filling the feature values ​​of the deleted pixel positions with any identical values.

[0085] In step S170, the enhanced feature map is passed through a classifier to obtain a classification loss function value. That is, in the technical solution architecture of the present application, feature extraction and feature-based classification are decoupled, that is, the first convolutional neural network focuses on feature extraction, and the classifier focuses on classification.

[0086] Specifically, in an embodiment of the present application, the classifier includes an encoder, which is used to encode the enhanced feature map to map the enhanced feature map into a label space to obtain a classification feature vector. In an embodiment of the present application, the encoder includes multiple convolutional layers and at least one fully connected layer, wherein the multiple convolutional layers are used to perform convolution processing on the enhanced feature map to obtain a deep feature map, and the one or more fully connected layers are used to further encode the deep feature map to obtain the classification feature vector. Furthermore, the classification feature vector obtained by encoding is passed through a Softmax classification function to obtain the classification loss function value, and the classification loss function value represents the probability that the classification result conforms to the true value.

[0087] Figure 5 The figure shows a flow chart of a neural network training method for detecting the light purification effect of organic waste gas according to an embodiment of the present application, in which the enhanced feature map is passed through a classifier to obtain a classification loss function value. Figure 5 As shown, in the embodiment of the present application, the enhanced feature map is passed through a classifier to obtain a classification loss function value, including: S310, passing the enhanced feature map through multiple convolutional layers to obtain a deep feature map; S320, passing the deep feature map through one or more fully connected layers to obtain a classification feature vector; S330, passing the classification feature vector through a Softmax classification function to obtain the classification loss function value

[0088] In step S180, based on the classification loss function value, the parameters of the first convolutional neural network and the classifier are updated. This is a conventional training process of a deep neural network, that is, the classification loss function is reduced by a preset step size, and the first convolutional neural network and the classifier are updated by directional propagation of gradient descent.

[0089] In summary, the training method of the neural network for detecting the light purification effect of organic waste gas based on the embodiment of the present application is explained. It uses computer vision technology based on artificial intelligence to detect the photocatalytic oxidation effect based on optical images, and when using a convolutional neural network to extract features from the image, the obtained feature map is subjected to feature screening processing so that the processed feature map can focus more on other image feature expressions other than particulate matter. In this way, the accuracy of light purification effect detection is improved.

[0090] According to another aspect of the present application, a method for detecting the optical purification effect of organic waste gas based on a deep neural network is also provided.

[0091] Figure 6 The figure illustrates a flow chart of a method for detecting the optical purification effect of organic waste gas based on a deep neural network according to an embodiment of the present application.

[0092] like Figure 6 As shown, according to the embodiment of the present application, the method for detecting the optical purification effect of organic waste gas based on a deep neural network includes: S410, obtaining an image to be detected, wherein the image to be detected is an image collected under the irradiation of a UV light source after the organic waste gas is photocatalytically oxidized; and, S420, inputting the image to be detected into a first convolutional neural network and a classifier trained according to the training method of a neural network for detecting the optical purification effect of organic waste gas as described above, to obtain a classification result, wherein the classification result is used to indicate whether the optical purification effect of the organic waste gas is qualified.

[0093] Exemplary Systems

[0094] Figure 7 The diagram illustrates a block diagram of a neural network training system for detecting the optical purification effect of organic waste gas according to an embodiment of the present application.

[0095] like Figure 7As shown, according to an embodiment of the present application, a neural network training system 700 for detecting the light purification effect of organic waste gas includes: a training image acquisition unit 710, which is used to acquire a plurality of first images of organic waste gas after photocatalytic oxidation under UV light source irradiation and a plurality of second images of organic waste gas after photocatalytic oxidation and detected to contain no organic gas but particulate impurities under the same UV light source irradiation; a training feature map generation unit 720, which is used to pass the plurality of first images obtained by the training image acquisition unit 710 through a first convolutional neural network to obtain a training feature map; a reference feature map generation unit 730, which is used to pass the plurality of second images obtained by the training image acquisition unit 710 through a second convolutional neural network to obtain a reference feature map, the first convolutional neural network and the second convolutional neural network having the same network structure; a difference map generation unit 740, which is used to calculate the training feature map generation unit 720. The obtained training feature map is differentiated by pixel position from the reference feature map obtained by the reference feature map generating unit 730 to obtain a difference map; a position mask matrix generating unit 750 is used to determine a set of positions in which the values ​​of each pixel position in the difference map obtained by the difference map generating unit 740 are less than a preset threshold value to obtain a position mask matrix; an enhanced feature map generating unit 760 is used to delete the values ​​of corresponding positions in the training feature map based on the position mask matrix obtained by the position mask matrix generating unit 750 to obtain an enhanced feature map; a classification loss function value calculating unit 770 is used to pass the enhanced feature map obtained by the enhanced feature map generating unit 760 through a classifier to obtain a classification loss function value; and a parameter updating unit 780 is used to update the parameters of the first convolutional neural network and the classifier based on the classification loss function value obtained by the classification loss function value calculating unit 770.

[0096] In one example, in the above-mentioned training system 700, the position mask matrix generating unit 750 further includes: in response to the value of the pixel position in the differential image being less than a preset threshold, setting the corresponding position to 0; in response to the value of the pixel position in the differential image being greater than or equal to a preset threshold, setting the corresponding position to 1; and arranging the encoding values ​​of the corresponding pixel positions in order to obtain the position mask matrix.

[0097] In one example, in the above-mentioned training system 700, the preset threshold is involved in the training process of the first convolutional neural network as a hyperparameter.

[0098] In one example, in the above-mentioned training system 700, the enhanced feature map generating unit 760 is further used to: perform a point multiplication of the position mask matrix and the training feature map at pixel positions to obtain the enhanced feature map.

[0099] In one example, in the above training system 700, as Figure 7 As shown, the classification loss function value calculation unit 770 includes: a convolution processing subunit 771, used to pass the enhanced feature map through multiple convolution layers to obtain a deep feature map; a fully connected processing subunit 772, used to pass the deep feature map through one or more fully connected layers to obtain a classification feature vector; and a loss function calculation subunit 773, used to pass the classification feature vector through a Softmax classification function to obtain the classification loss function value.

[0100] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the training system 700 have been described in detail above. Figures 1 to 5 The training method of the neural network for detecting the optical purification effect of organic waste gas has been described in detail, and therefore, its repeated description will be omitted.

[0101] As described above, the training system 700 according to the embodiment of the present application can be implemented in various terminal devices, such as a server for detecting the light purification effect of organic waste gas. In one example, the training system 700 according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the training system 700 can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the training system 700 can also be one of the many hardware modules of the terminal device.

[0102] Alternatively, in another example, the training system 700 and the terminal device may also be separate devices, and the training system 700 may be connected to the terminal device via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0103] According to another aspect of the present application, a light purification effect detection system for organic waste gas based on a deep neural network is also proposed.

[0104] Fig. 9 The figure shows a block diagram of a light purification effect detection system for organic waste gas based on a deep neural network according to an embodiment of the present application. Fig. 9As shown, according to an embodiment of the present application, a system 800 for detecting the optical purification effect of organic waste gas based on a deep neural network includes: an image acquisition unit 810 for acquiring an image to be detected, wherein the image to be detected is an image collected after the organic waste gas is photocatalytically oxidized under the irradiation of a UV light source; and a classification result generation unit 820, which is used to input the image to be detected obtained by the image acquisition unit 810 into a first convolutional neural network and a classifier trained according to the training method of a neural network for detecting the optical purification effect of organic waste gas as described above, so as to obtain a classification result, wherein the classification result is used to indicate whether the optical purification effect of the organic waste gas is qualified.

[0105] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the control system 900 have been described in detail above. Figure 6 The description of the light purification effect detection method for organic waste gas based on deep neural network has been introduced in detail, and therefore, its repeated description will be omitted.

[0106] As described above, the detection system 900 according to the embodiment of the present application can be implemented in various terminal devices, such as a server for detecting the light purification effect of organic waste gas. In one example, the detection system 900 according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the detection system 900 can be a software module in the operating system of the terminal device, or it can be an application developed for the terminal device; of course, the detection system 900 can also be one of the many hardware modules of the terminal device.

[0107] Alternatively, in another example, the detection system 900 and the terminal device may also be separate devices, and the detection system 900 may be connected to the terminal device via a wired and / or wireless network, and transmit the interactive information in accordance with an agreed data format.

[0108] Exemplary Electronic Devices

[0109] Below, reference Fig.10 To describe an electronic device according to an embodiment of the present application.

[0110] Fig.10 A block diagram of an electronic device according to an embodiment of the present application is illustrated.

[0111] like Fig.10 As shown, the electronic device 10 includes one or more processors 11 and a memory 12 .

[0112] The processor 11 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.

[0113] The memory 12 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 11 may run the program instructions to implement the training method of a neural network for detecting the light purification effect of organic waste gas in each embodiment of the present application described above, or a method for detecting the light purification effect of organic waste gas based on a deep neural network and / or other desired functions. Various contents such as a first image, a second image, a classification result, etc. may also be stored in the computer-readable storage medium.

[0114] In one example, the electronic device 10 may further include: an input device 13 and an output device 14, and these components are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0115] The input device 13 may include, for example, a keyboard, a mouse, etc.

[0116] The output device 14 can output various information to the outside, including classification results, etc. The output device 14 can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, and the like.

[0117] Of course, to simplify, Fig.10 Only some of the components related to the present application in the electronic device 10 are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, according to specific application situations, the electronic device 10 may also include any other appropriate components.

[0118] Exemplary computer program products and computer-readable storage media

[0119] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the training method of a neural network for detecting the optical purification effect of organic waste gas according to various embodiments of the present application described in the above "Exemplary Methods" section of this specification, or the steps in the method for detecting the optical purification effect of organic waste gas based on a deep neural network.

[0120] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0121] In addition, an embodiment of the present application may also be a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, causes the processor to execute the training method of a neural network for detecting the optical purification effect of organic waste gas according to various embodiments of the present application described in the above “Exemplary Method” section of this specification, or the steps in a method for detecting the optical purification effect of organic waste gas based on a deep neural network.

[0122] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0123] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, not for limitation, and the above details do not limit the present application to being implemented by adopting the above specific details.

[0124] The block diagrams of the devices, apparatuses, equipment, and systems involved in this application are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The words "such as" used here refer to the phrase "such as but not limited to", and can be used interchangeably with them.

[0125] It should also be noted that in the apparatus, device and method of the present application, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present application.

[0126] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0127] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A neural network training method for detecting the light purification effect of organic waste gas, characterized in that: include: Acquire a plurality of first images of the organic waste gas after photocatalytic oxidation under the irradiation of a UV light source and a plurality of second images of the organic waste gas after photocatalytic oxidation and detected to contain no organic gas but particulate impurities under the irradiation of the same UV light source; Passing the plurality of first images through a first convolutional neural network to obtain a training feature map; Passing the plurality of second images through a second convolutional neural network to obtain a reference feature map, wherein the first convolutional neural network and the second convolutional neural network have the same network structure; Calculating the pixel-wise difference between the training feature map and the reference feature map to obtain a difference map; Determine a position set where the value of each pixel position in the difference image is less than a preset threshold to obtain a position mask matrix; Based on the position mask matrix, deleting the value of the corresponding position in the training feature map to obtain an enhanced feature map; Passing the enhanced feature map through a classifier to obtain a classification loss function value; as well as Based on the classification loss function value, updating parameters of the first convolutional neural network and the classifier; Wherein, determining a position set whose value of each pixel position in the difference image is less than a preset threshold to obtain a position mask matrix includes: In response to a value of a pixel position in the differential image being less than a preset threshold, setting the corresponding position to 0; In response to a value of a pixel position in the difference image being greater than or equal to a preset threshold, setting the corresponding position to 1; and The encoding values ​​corresponding to the pixel positions are arranged in order to obtain the position mask matrix.

2. The neural network training method for detecting the light purification effect of organic waste gas according to claim 1, wherein: The preset threshold is used as a hyperparameter in the training process of the first convolutional neural network.

3. The neural network training method for detecting the light purification effect of organic waste gas according to claim 1, wherein: Deleting the value of the corresponding position in the training feature map based on the position mask matrix to obtain an enhanced feature map, comprising: The position mask matrix is ​​multiplied by the training feature map at pixel positions to obtain the enhanced feature map.

4. The neural network training method for detecting the light purification effect of organic waste gas according to claim 1, wherein: Passing the enhanced feature map through a classifier to obtain a classification loss function value includes: Passing the enhanced feature map through multiple convolutional layers to obtain a deep feature map; Passing the deep feature map through one or more fully connected layers to obtain a classification feature vector; and The classification feature vector is passed through the Softmax classification function to obtain the classification loss function value.

5. The neural network training method for detecting the light purification effect of organic waste gas according to claim 1, wherein: The first convolutional neural network and the second convolutional neural network are deep residual networks.

6. A method for detecting the light purification effect of organic waste gas based on a deep neural network, characterized in that: include: Acquire an image to be detected, wherein the image to be detected is an image collected under UV light source after the organic waste gas is photocatalytically oxidized; as well as The image to be detected is input into the first convolutional neural network and classifier trained according to the neural network training method for detecting the optical purification effect of organic waste gas as described in any one of claims 1 to 5 to obtain a classification result, and the classification result is used to indicate whether the optical purification effect of the organic waste gas is qualified.

7. A neural network training system for detecting the optical purification effect of organic waste gas, characterized in that: include: A training image acquisition unit, used to acquire a plurality of first images of the organic waste gas after photocatalytic oxidation under the irradiation of a UV light source and a plurality of second images of the organic waste gas after photocatalytic oxidation and detected to contain no organic gas but particulate impurities under the irradiation of the same UV light source; a training feature map generating unit, configured to pass the plurality of first images obtained by the training image obtaining unit through a first convolutional neural network to obtain a training feature map; a reference feature map generating unit, configured to pass the plurality of second images obtained by the training image obtaining unit through a second convolutional neural network to obtain a reference feature map, wherein the first convolutional neural network and the second convolutional neural network have the same network structure; a difference map generating unit, configured to calculate a pixel-wise difference between the training feature map obtained by the training feature map generating unit and the reference feature map obtained by the reference feature map generating unit, so as to obtain a difference map; A position mask matrix generating unit, used to determine a position set where the value of each pixel position in the difference image obtained by the difference image generating unit is less than a preset threshold, so as to obtain a position mask matrix; an enhanced feature map generating unit, configured to delete the value of the corresponding position in the training feature map based on the position mask matrix obtained by the position mask matrix generating unit to obtain an enhanced feature map; a classification loss function value calculation unit, configured to pass the enhanced feature map obtained by the enhanced feature map generation unit through a classifier to obtain a classification loss function value; as well as A parameter updating unit, configured to update parameters of the first convolutional neural network and the classifier based on the classification loss function value obtained by the classification loss function value calculating unit; Wherein, the position mask matrix generating unit comprises: In response to a value of a pixel position in the differential image being less than a preset threshold, setting the corresponding position to 0; In response to a value of a pixel position in the difference image being greater than or equal to a preset threshold, setting the corresponding position to 1; and The encoding values ​​corresponding to the pixel positions are arranged in order to obtain the position mask matrix.

8. A light purification effect detection system for organic waste gas based on deep neural network, characterized in that: include: The image acquisition unit to be detected is used to acquire the image to be detected, wherein the image to be detected is an image collected after the organic waste gas is photocatalytically oxidized under the irradiation of a UV light source; as well as A classification result generating unit is used to input the image to be detected obtained by the image to be detected acquiring unit into a first convolutional neural network and a classifier trained according to the training method of a neural network for detecting the optical purification effect of organic waste gas as described in any one of claims 1 to 5, so as to obtain a classification result, wherein the classification result is used to indicate whether the optical purification effect of the organic waste gas is qualified.

9. An electronic device, comprising: processor; as well as A memory, in which computer program instructions are stored, and when the computer program instructions are executed by the processor, the processor executes the training method of a neural network for detecting the light purification effect of organic waste gas as described in any one of claims 1 to 5, or the method for detecting the light purification effect of organic waste gas based on a deep neural network as described in claim 6.