Dedusting method and system based on artificial intelligence

Abstract

The invention relates to a dust removal method and system based on artificial intelligence, and belongs to the technical field of intelligent dust removal. The method comprises the steps of obtaining a historical dust removal area image set, constructing a dust recognition data set, obtaining a dust detection model through lightweight NanoDet network training and testing, further generating a dust area data set, obtaining a dust sub-area enhanced image set through pixel-level segmentation and contrast enhancement, and constructing a dust concentration data set; obtaining a dust concentration prediction model through lightweight residual network training and testing according to the dust concentration data set; obtaining a real-time dust removal area image, obtaining a dust removal sub-area dust concentration set through the dust detection model and the dust concentration prediction model, further obtaining the dust concentration of a target area through dust concentration weighted averaging, dividing dust concentration grades through a dust concentration threshold value, and adjusting a dust removal strategy. Therefore, the dust concentration of the target area is accurately recognized, the dust removal mode is dynamically adjusted, and the dust removal efficiency and adaptability are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent dust removal, and in particular relates to a dust removal method and system based on artificial intelligence. Background Art

[0002] In the field of industrial production, dust pollution is a common and serious problem. It not only damages the environment, but also seriously threatens the health of workers. Traditional dust removal methods mainly rely on fixed parameter settings and manual intervention. The limitation of this method is that the fixed dust removal parameters lead to poor dust removal adaptability and cannot dynamically adjust the dust removal strategy according to actual environmental changes, resulting in low dust removal efficiency.

[0003] Existing dust removal technologies often fail to effectively solve the following technical problems: lack of real-time and accurate dust concentration detection methods, low data processing efficiency, difficulty in quickly and accurately identifying dust concentrations in target areas, and inability to meet the needs of real-time monitoring and adjustment of dust removal strategies, which in turn affects the dust removal effect. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention proposes a dust removal method and system based on artificial intelligence.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A dust removal method based on artificial intelligence, comprising:

[0007] Acquire a historical dust removal area image set, and construct a dust recognition data set through data enhancement based on the historical dust removal area image set;

[0008] A dust detection model is obtained by training and testing the dust recognition data set through a lightweight NanoDet network, and a dust region data set is obtained through the dust detection model according to the dust recognition data set, wherein the dust region data set is a dust recognition data set with a dust sub-region detection frame;

[0009] A dust sub-region image set is obtained by separating the dust region data set through pixel-level segmentation technology, and a dust sub-region enhanced image set is obtained by contrast enhancement based on the dust sub-region image set;

[0010] A dust concentration dataset is constructed based on the dust sub-region enhanced image set by annotating the dust concentration values, and a dust concentration prediction model is obtained by training and testing a lightweight residual network based on the dust concentration dataset;

[0011] Acquire a real-time dust removal area image, obtain a dust removal sub-area enhanced image set based on the real-time dust removal area image through the dust detection model, and obtain a dust removal sub-area dust concentration set based on the dust removal sub-area enhanced image set through the dust concentration prediction model;

[0012] The dust concentration in the target area is obtained by weighted average of the dust concentration according to the dust concentration set of the dust removal sub-area, the dust concentration level is obtained by dividing the dust concentration threshold according to the dust concentration in the target area, and the dust removal strategy is adjusted according to the dust concentration level.

[0013] Preferably, the dust detection model obtained by training and testing the dust identification data set through a lightweight NanoDet network includes:

[0014] Based on the backbone network MobileNetV3, a lightweight NanoDet network is constructed through feature aggregation;

[0015] Training and testing the lightweight NanoDet network according to the dust identification dataset to obtain an initial dust detection model;

[0016] According to the dust detection initial model, the dust detection model is obtained by evaluating the dust detection GLOU loss function and the average precision of dust detection.

[0017] Preferably, constructing a lightweight NanoDet network by feature aggregation based on the backbone network MobileNetV3 includes:

[0018] According to the dust recognition data set, multi-level dust image features are extracted through the MobileNetV3 backbone network;

[0019] According to the multi-level dust image features, the dust image features are obtained by fusing the features of the lightweight path aggregation network;

[0020] According to the dust image features, a dust sub-region detection frame is obtained by detecting with the GFiocalV2 anchor-free lightweight detection head, and each dust sub-region detection frame is accompanied by a confidence score.

[0021] Preferably, the calculation process of the average accuracy of dust detection is:

[0022] Sort the dust sub-region detection frames in descending order according to the confidence scores to obtain a list of dust sub-region detection frames in order;

[0023] According to the ordered dust sub-region detection frame list, an average dust detection accuracy is calculated by matching the dust real annotations;

[0024] The mathematical expression of the average accuracy of dust detection is:

[0025]

[0026] R=[0,0.1,0.2,…,1.0]

[0027] Among them, AP is the average precision of dust detection, R is the dust detection recall rate threshold set, r is the dust detection recall rate threshold, P is the dust detection precision, and recall is the dust detection recall rate.

[0028] Preferably, the dust concentration prediction model obtained by training and testing a lightweight residual network according to the dust concentration data set includes:

[0029] According to the residual network resnet50, a lightweight residual network is constructed by replacing convolution blocks, wherein the convolution blocks are replaced by replacing standard convolutions in the residual network resnet50 with depthwise separable convolutions;

[0030] The dust concentration prediction model is obtained by training and testing the lightweight residual network through Adam optimization parameters according to the dust concentration data set.

[0031] Preferably, obtaining the dust concentration of the target area by weighted average of dust concentrations according to the dust concentration set of the dust removal sub-areas includes:

[0032] The dust concentration set of the dust removal sub-area is obtained by weighted calculation of dust concentration according to the dust concentration set of the dust removal sub-area. The mathematical expression of the weighted calculation of dust concentration is:

[0033] C=a×sC

[0034] Among them, c is the dust weighted concentration in the dust removal sub-area, a is the dust weighted coefficient in the dust removal sub-area, and sc is the dust concentration in the dust removal sub-area;

[0035] The dust concentration in the target area is obtained by weighted average of the dust concentration according to the dust weighted concentration set of the dust removal sub-area. The mathematical expression of the weighted average of the dust concentration is:

[0036]

[0037] Among them, C is the dust concentration in the target area, c is the weighted dust concentration in the dust removal sub-area, and n is the number of dust removal sub-areas.

[0038] Preferably, the adjusting the dust removal strategy according to the dust concentration level includes:

[0039] The dust concentration levels include high dust concentration level, medium dust concentration level, and low dust concentration level;

[0040] When the dust concentration level is a high dust concentration level, a high-intensity dust removal mode is started;

[0041] When the dust concentration level is a medium dust concentration level, a medium intensity dust removal mode is started;

[0042] When the dust concentration level is a low dust concentration level, a low-intensity dust removal mode is started.

[0043] An artificial intelligence-based dust removal system using the artificial intelligence-based dust removal method, comprising a dust removal image processing module, a dust area detection module, a dust concentration detection module, and a dust removal mode control module;

[0044] The dust removal image processing module is used to obtain the dust removal area image and perform image denoising preprocessing to obtain the dust removal area standard image;

[0045] The dust area detection module is used to detect the dust removal area standard image to obtain a dust sub-area detection frame and its confidence score;

[0046] The dust concentration detection module is used to detect the dust concentration in the dust removal sub-area and calculate the dust concentration in the target area;

[0047] The dust removal mode control module is used to obtain the dust concentration level according to the dust concentration in the target area and adjust the dust removal strategy.

[0048] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned artificial intelligence-based dust removal method when executing the program.

[0049] A storage medium containing computer executable instructions, which are used to perform the above-mentioned artificial intelligence-based dust removal method when executed by a computer processor.

[0050] The beneficial effects of the present invention are:

[0051] (1) The dust detection model and dust concentration prediction model are used to identify and detect the dust concentration in the target area of ​​the real-time dust removal area image, and then the dust concentration level is divided according to the dust concentration threshold and the dust removal strategy is adjusted. The dust concentration in the target area can be accurately identified, the dust removal mode can be dynamically adjusted, and the dust removal efficiency and adaptability can be improved.

[0052] (2) Multi-level dust image features are extracted through the MobileNetV3 backbone network and the dust image features are obtained through lightweight path aggregation network feature fusion. Then, the dust sub-region detection frame is obtained through GFicalV2 anchor-free lightweight detection head detection. The dust sub-region detection frame and its confidence score can be generated quickly and accurately, effectively improving the efficiency and accuracy of the dust detection model.

[0053] (3) By constructing a lightweight residual network, the standard convolution in the residual network resnet50 is replaced with a depthwise separable convolution, which reduces the number of parameters of the dust concentration prediction model and improves the running speed of the dust concentration prediction model.

[0054] (4) The weighted concentration of the dust concentration set in the dust removal sub-area is calculated to form a dust weighted concentration set in the dust removal sub-area, and then the dust concentration in the target area is obtained by weighted average of the dust. The area ratio of the dust removal sub-area is taken into consideration, ensuring the accuracy and rationality of the dust concentration calculation in the overall target area. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0056] Figure 1 The figure is a schematic flow chart of a dust removal method based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0057] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0058] See also Figure 1 , an artificial intelligence-based dust removal method, comprising:

[0059] S1: Acquire a historical dust removal area image set, and construct a dust recognition data set through data enhancement based on the historical dust removal area image set;

[0060] S2: A dust detection model is obtained by training and testing the dust recognition data set through a lightweight NanoDet network, and a dust region data set is obtained through the dust detection model according to the dust recognition data set, wherein the dust region data set is a dust recognition data set with a dust sub-region detection frame;

[0061] S3: Separating the dust region data set by pixel-level segmentation technology to obtain a dust sub-region image set, and obtaining a dust sub-region enhanced image set by contrast enhancement according to the dust sub-region image set;

[0062] S4: constructing a dust concentration dataset by annotating dust concentration values ​​according to the dust sub-region enhanced image set, and obtaining a dust concentration prediction model by training and testing a lightweight residual network according to the dust concentration dataset;

[0063] S5: acquiring a real-time dust removal area image, obtaining a dust removal sub-area enhanced image set based on the real-time dust removal area image by using the dust detection model, and obtaining a dust removal sub-area dust concentration set based on the dust removal sub-area enhanced image set by using the dust concentration prediction model;

[0064] S6: obtaining the dust concentration of the target area by weighted average of the dust concentration set of the dust removal sub-areas, obtaining the dust concentration level by dividing the dust concentration threshold value according to the dust concentration of the target area, and adjusting the dust removal strategy according to the dust concentration level.

[0065] In this embodiment, constructing a dust concentration dataset through data enhancement based on the historical dust removal area images can be specifically implemented by the following steps:

[0066] S101: obtaining a standard image of the historical dust removal area by performing image denoising according to the image of the historical dust removal area;

[0067] S102: obtaining a historical dust removal area standard annotated image by annotating dust removal area data according to the historical dust removal area standard image, wherein the dust removal area data annotating includes annotating a dust sub-area target frame by using a LabelImg annotation tool according to the historical dust removal area standard image;

[0068] S103: obtaining a dust concentration dataset by data enhancement based on the standard annotated images of the historical dust removal areas, wherein the data enhancement includes random rotation and horizontal flipping.

[0069] In this embodiment, the dust detection model obtained by training and testing the lightweight NanoDet network according to the dust identification data set can be implemented by the following steps:

[0070] S201: Construct a lightweight NanoDet network through feature aggregation based on the backbone network MobileNetV3;

[0071] The step S201 can be specifically implemented by the following steps:

[0072] S201-1: extracting multi-level dust image features through the MobileNetV3 backbone network according to the dust recognition dataset;

[0073] S201-2: Obtain dust image features through lightweight path convergence network feature fusion according to the multi-level dust image features;

[0074] S201-3: A dust sub-region detection frame is obtained based on the dust image features through GFiocalV2 anchor-free lightweight detection head detection, and each dust sub-region detection frame is accompanied by a confidence score.

[0075] S202: training and testing the lightweight NanoDet network according to the dust recognition data set to obtain an initial dust detection model;

[0076] Specifically, the dust recognition data set is divided into a dust recognition training set and a dust recognition test set in a ratio of 8:2.

[0077] S203: obtaining a dust detection model according to the dust detection initial model through the dust detection GLOU loss function and the dust detection average accuracy evaluation;

[0078] The mathematical expression of the GLOU loss function is:

[0079]

[0080]

[0081] Among them, L is the intersection-over-union ratio of dust detection, A is the dust sub-region detection box, B is the actual box of the dust sub-region, G is the GLOU loss function of dust detection, C is the minimum area containing the dust sub-region detection box and the actual box of the dust sub-region, and U is the union of the dust sub-region detection box and the actual box of the dust sub-region.

[0082] The calculation steps of the average accuracy of dust detection are as follows:

[0083] S203-1: Sort the dust sub-region detection frames in descending order according to the confidence scores to obtain a list of dust sub-region detection frames in order;

[0084] S203-2: Calculate the average dust detection accuracy through dust real annotation matching based on the ordered dust sub-region detection frame list.

[0085] In this embodiment, the dust detection intersection-and-union ratio threshold is set to 0.75. When the dust detection intersection-and-union ratio is greater than the dust detection intersection-and-union ratio threshold, the dust sub-region detection frame is detected correctly.

[0086] Specifically, the dust detection intersection-and-union ratio is calculated for each dust sub-region detection frame in the ordered dust sub-region detection frame list, and the dust detection precision and dust detection recall are obtained through statistical calculation according to the dust detection intersection-and-union ratio.

[0087] The mathematical expressions of the dust detection accuracy and the dust detection recall are:

[0088]

[0089]

[0090] Among them, P is the dust detection precision, recall is the dust detection recall rate, TP is the number of dust sub-region target frames that are correctly detected as dust sub-region detection frames, FP is the number of dust sub-region target frames that are not dust sub-region target frames but are mistakenly detected as dust sub-region detection frames, and FN is the number of dust sub-region target frames that are not detected.

[0091] The dust detection average accuracy is calculated by 11-point interpolation method according to the dust detection accuracy. The mathematical expression of the dust detection average accuracy is:

[0092]

[0093] R=[0,0.1,0.2,…,1.0]

[0094] Among them, AP is the average precision of dust detection, R is the dust detection recall rate threshold set, r is the dust detection recall rate threshold, P is the dust detection precision, and recall is the dust detection recall rate.

[0095] When the GLOU loss function value of the initial dust detection model does not decrease significantly within 10 consecutive iteration cycles or the average dust detection accuracy of the initial dust detection model reaches 90% or the training iteration of the initial dust detection model reaches a predetermined maximum number of iterations, the training process will stop, and the model generated in the current iteration cycle is the dust detection model.

[0096] In this embodiment, the dust detection model can identify multiple dust sub-regions in the dust removal area image and generate multiple corresponding dust sub-region detection frames. At the same time, the dust sub-region detection frames with high overlap are removed through non-maximum suppression, and only the dust sub-region detection frames with the highest confidence scores are retained.

[0097] In this embodiment, obtaining the dust area dataset through the dust detection model according to the dust identification dataset can be specifically implemented by the following steps:

[0098] Obtaining a dust sub-area set through the dust detection model according to the historical dust removal area images;

[0099] The historical dust removal area images in the dust recognition data set are traversed to obtain a dust area data set, wherein the dust sub-area set is the historical dust removal area images with dust sub-area detection frames.

[0100] In this embodiment, the dust sub-region image set is obtained by separating the dust region data set by pixel-level segmentation technology, which can be implemented by the following steps:

[0101] S301: obtaining a dust sub-region image by performing pixel-level dust region segmentation by PSPNet according to the dust sub-region set;

[0102] S302: Traverse the dust region data set to obtain a dust sub-region image set.

[0103] In this embodiment, the dust concentration prediction model is obtained by training and testing the dust concentration data set through a lightweight residual network, which can be implemented by the following steps:

[0104] S401: constructing a lightweight residual network by replacing convolution blocks according to the residual network resnet50, wherein the convolution block replacement is to replace the standard convolution in the residual network resnet50 with a depth-separable convolution;

[0105] S402: According to the dust concentration data set, the lightweight residual network is trained and tested by Adam optimization parameters to obtain the dust concentration prediction model.

[0106] In this embodiment, obtaining the dust concentration of the target area by weighted average of the dust concentrations according to the dust concentration set of the dust removal sub-areas can be specifically implemented by the following steps:

[0107] S601: Obtain a dust weighted concentration set of the dust removal sub-area by performing dust concentration weighted calculation according to the dust concentration set of the dust removal sub-area. The mathematical expression of the dust concentration weighted calculation is:

[0108] C=a×sc

[0109] Among them, c is the dust weighted concentration in the dust removal sub-area, a is the dust weighted coefficient in the dust removal sub-area, and sc is the dust concentration in the dust removal sub-area;

[0110] Specifically, the dust weighting coefficient of the dust removal sub-area is the ratio of the area of ​​the dust removal sub-area to the area of ​​the target area.

[0111] S602: Obtain the dust concentration of the target area by weighted average of the dust concentration according to the dust weighted concentration set of the dust removal sub-area. The mathematical expression of the weighted average of the dust concentration is:

[0112]

[0113] Among them, C is the dust concentration in the target area, c is the weighted dust concentration in the dust removal sub-area, and n is the number of dust removal sub-areas.

[0114] In this embodiment, the dust concentration level is obtained by dividing the dust concentration in the target area by a dust concentration threshold, and adjusting the dust removal strategy according to the dust concentration level includes:

[0115] Setting a dust concentration threshold, and obtaining a dust concentration level of the target area according to the dust concentration threshold, wherein the dust concentration level includes a high dust concentration level, a medium dust concentration level, and a low dust concentration level;

[0116] When the dust concentration level is a high dust concentration level, a high-intensity dust removal mode is started;

[0117] When the dust concentration level is a medium dust concentration level, a medium intensity dust removal mode is started;

[0118] When the dust concentration level is a low dust concentration level, a low-intensity dust removal mode is started.

[0119] An artificial intelligence-based dust removal system includes a dust removal image processing module, a dust area detection module, a dust concentration detection module, and a dust removal mode control module;

[0120] The dust removal image processing module is used to obtain the dust removal area image and perform image denoising preprocessing to obtain the dust removal area standard image;

[0121] The dust area detection module is used to detect the dust removal area standard image to obtain a dust sub-area detection frame and its confidence score;

[0122] The dust concentration detection module is used to detect the dust concentration in the dust removal sub-area and calculate the dust concentration in the target area;

[0123] The dust removal mode control module is used to obtain the dust concentration level according to the dust concentration in the target area and adjust the dust removal strategy.

[0124] The computer storage medium of the embodiment of the present invention may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer 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. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device.

[0125] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0126] The program code included on the computer readable medium can be transmitted with any appropriate medium, including but not limited to wireless, electric wire, optical cable, RF, etc., or any suitable combination of the above. The computer program code for performing the operation of the present invention can be written in one or more programming languages ​​or their combinations, and the programming language includes object-oriented programming languages-such as Java, Smalltalk, C++, and also includes conventional procedural programming languages-such as "C" language or similar programming languages. The program code can be executed completely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on the remote computer, or completely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0127] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A dust removal method based on artificial intelligence, characterized in that: include: Acquire a historical dust removal area image set, and construct a dust recognition data set through data enhancement based on the historical dust removal area image set; A dust detection model is obtained by training and testing the dust recognition data set through a lightweight NanoDet network, and a dust region data set is obtained through the dust detection model according to the dust recognition data set, wherein the dust region data set is a dust recognition data set with a dust sub-region detection frame; A dust sub-region image set is obtained by separating the dust region data set through pixel-level segmentation technology, and a dust sub-region enhanced image set is obtained by contrast enhancement based on the dust sub-region image set; A dust concentration dataset is constructed based on the dust sub-region enhanced image set by annotating the dust concentration values, and a dust concentration prediction model is obtained by training and testing a lightweight residual network based on the dust concentration dataset; Acquire a real-time dust removal area image, obtain a dust removal sub-area enhanced image set based on the real-time dust removal area image through the dust detection model, and obtain a dust removal sub-area dust concentration set based on the dust removal sub-area enhanced image set through the dust concentration prediction model; The dust concentration in the target area is obtained by weighted average of the dust concentration according to the dust concentration set of the dust removal sub-area, the dust concentration level is obtained by dividing the dust concentration threshold according to the dust concentration in the target area, and the dust removal strategy is adjusted according to the dust concentration level.

2. The dust removal method based on artificial intelligence according to claim 1, characterized in that: The dust detection model obtained by training and testing the lightweight NanoDet network according to the dust identification data set includes: Based on the backbone network MobileNetV3, a lightweight NanoDet network is constructed through feature aggregation; Training and testing the lightweight NanoDet network according to the dust identification dataset to obtain an initial dust detection model; According to the dust detection initial model, the dust detection model is obtained by evaluating the dust detection GLOU loss function and the average precision of dust detection.

3. The dust removal method based on artificial intelligence according to claim 2, characterized in that: The lightweight NanoDet network constructed by feature aggregation based on the backbone network MobileNetV3 includes: According to the dust recognition data set, multi-level dust image features are extracted through the MobileNetV3 backbone network; According to the multi-level dust image features, the dust image features are obtained by fusing the features of the lightweight path aggregation network; According to the dust image features, a dust sub-region detection frame is obtained by detecting with the GFiocalV2 anchor-free lightweight detection head, and each dust sub-region detection frame is accompanied by a confidence score.

4. The dust removal method based on artificial intelligence according to claim 2, characterized in that: The calculation process of the average accuracy of dust detection is as follows: Sort the dust sub-region detection frames in descending order according to the confidence scores to obtain a list of dust sub-region detection frames in order; According to the ordered dust sub-region detection frame list, an average dust detection accuracy is calculated by matching the dust real annotations; The mathematical expression of the average accuracy of dust detection is: R＝[0,0.10.2,…,1.0] Among them, AP is the average precision of dust detection, R is the dust detection recall rate threshold set, r is the dust detection recall rate threshold, P is the dust detection precision, and recall is the dust detection recall rate.

5. The dust removal method based on artificial intelligence according to claim 1, characterized in that: The dust concentration prediction model obtained by training and testing the dust concentration data set through a lightweight residual network includes: According to the residual network resnet50, a lightweight residual network is constructed by replacing convolution blocks, wherein the convolution blocks are replaced by replacing standard convolutions in the residual network resnet50 with depthwise separable convolutions; According to the dust concentration data set, Ada m The light residual network is trained and tested by optimizing parameters to obtain the dust concentration prediction model.

6. The dust removal method based on artificial intelligence according to claim 1, characterized in that: The step of obtaining the dust concentration of the target area by weighted average of the dust concentrations according to the dust concentration set of the dust removal sub-areas includes: The dust concentration set of the dust removal sub-area is obtained by weighted calculation of dust concentration according to the dust concentration set of the dust removal sub-area. The mathematical expression of the weighted calculation of dust concentration is: C=a×sc Among them, c is the dust weighted concentration in the dust removal sub-area, a is the dust weighted coefficient in the dust removal sub-area, sc is the dust concentration in the dust removal sub-area; The dust concentration in the target area is obtained by weighted average of the dust concentration according to the dust weighted concentration set of the dust removal sub-area. The mathematical expression of the weighted average of the dust concentration is: Among them, C is the dust concentration in the target area, c is the weighted dust concentration in the dust removal sub-area, and n is the number of dust removal sub-areas.

7. The dust removal method based on artificial intelligence according to claim 1, characterized in that: The dust removal strategy is adjusted according to the dust concentration level, including: The dust concentration levels include high dust concentration level, medium dust concentration level, and low dust concentration level; When the dust concentration level is a high dust concentration level, a high-intensity dust removal mode is started; When the dust concentration level is a medium dust concentration level, a medium intensity dust removal mode is started; When the dust concentration level is a low dust concentration level, a low-intensity dust removal mode is started.

8. An artificial intelligence-based dust removal system using the artificial intelligence-based dust removal method of claim 1, comprising a dust removal image processing module, a dust area detection module, a dust concentration detection module, and a dust removal mode control module, characterized in that: The dust removal image processing module is used to obtain the dust removal area image and perform image denoising preprocessing to obtain the dust removal area standard image; The dust area detection module is used to detect the dust removal area standard image to obtain a dust sub-area detection frame and its confidence score; The dust concentration detection module is used to detect the dust concentration in the dust removal sub-area and calculate the dust concentration in the target area; The dust removal mode control module is used to obtain the dust concentration level according to the dust concentration in the target area and adjust the dust removal strategy.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the artificial intelligence-based dust removal method as described in any one of claims 1-7 is implemented.

10. A storage medium containing computer executable instructions, characterized in that: The computer executable instructions are used to execute the artificial intelligence-based dust removal method as described in any one of claims 1-7 when executed by a computer processor.

Images