Target detection early warning system and method for construction dust raising and prevention and control measures

Abstract

The invention relates to the technical field of dust identification and prevention, and discloses a target detection early warning system and method for construction dust and prevention measures, and the method comprises the steps: obtaining a construction site image data set with marking information from a historical database; constructing a target detection model of construction dust and prevention and control measures; the data set is transmitted into a target detection model of construction dust raising and prevention and control measures, and iterative training is carried out on the model; and acquiring image data corresponding to the current to-be-detected position, transmitting the image data as input to the trained target detection model of the construction dust and the prevention and control measures, outputting a target detection result of the construction dust and the prevention and control measures corresponding to the to-be-detected position, and sending the target detection result to the client.

Description

Technical Field

[0001] The present invention relates to the technical field of dust identification and prevention, and in particular to a target detection and early warning system and method for construction dust and prevention measures. Background Art

[0002] Construction sites are the main source of urban air pollution, and multiple scientific studies have shown that exposure to construction dust increases the risk of various respiratory diseases, cardiovascular diseases, and cancer. Compared with the general population, frontline workers and managers on construction sites face significantly increased health risks. In addition, since tiny particles in the air can easily spread to nearby roads and buildings, people around construction sites are also at a higher risk of injury. Therefore, it is very important to conduct continuous real-time monitoring of construction dust on the site and take effective preventive measures to reduce exposure.

[0003] Although construction dust is an important source of particulate matter pollution in the air, there has been a lack of research on the detection and monitoring of construction dust. There are two main methods for dust detection. One is manual visual detection, but this method will cause a great waste of human resources and high costs. The other is fixed-point dust concentration detection, which is easily affected by distance, wind direction, and obstructions, and the detection accuracy is unstable. Summary of the invention

[0004] One of the purposes of the present invention is to provide a target detection and early warning system and method for construction dust and prevention measures, which can achieve accurate detection of construction dust and greatly improve the accuracy and reliability of construction dust detection.

[0005] In order to achieve the above object, a target detection and early warning method for construction dust and prevention measures is provided, comprising the following steps:

[0006] S1, obtain the construction site image dataset with annotation information from the historical database;

[0007] S2. Construct a target detection model for construction dust and prevention measures; the target detection model for construction dust and prevention measures includes a Backbone module and a Head module; the Backbone module includes a CBS structure, an ELAN module, an MPConv structure, an MP_DCN structure, and an ELAN_DCN structure; the MP_DCN structure is formed by incorporating a DCDv2 structure into a final MP structure, and the ELAN_DCN structure is formed by incorporating a DCDv2 structure into a final ELAN structure;

[0008] S3, transferring the construction site image dataset with labeled information to the target detection model of construction dust and prevention measures, and iteratively training the model;

[0009] S4. Access the on-site monitoring system of the construction site, obtain the image data corresponding to the current position to be measured, and transmit the image data as input to the trained target detection model of construction dust and prevention measures, output the target detection result of construction dust and prevention measures corresponding to the position to be measured, and send it to the client.

[0010] Principle and effect of this solution: In this solution, the first step is to obtain a construction site image dataset with annotated information from the database, and classify the data in the dataset by annotating the construction site image dataset, thereby improving the efficiency of subsequent model training. At the same time, it can provide effective and accurate data support for subsequent model training and improve the efficiency of model training.

[0011] Then, a corresponding target detection model for construction dust and prevention measures is constructed, and the target detection model for construction dust and prevention measures includes a Backbone module and a Head module; the Backbone module includes a CBS structure, an ELAN module, an MPConv structure, an MP_DCN structure and an ELAN_DCN structure; the MP_DCN structure is formed by incorporating the DCDv2 structure into the final MP structure, and the ELAN_DCN structure is formed by incorporating the DCDv2 structure into the final ELAN structure; in this scheme, the DCDv2 structure is used in the Backbone module to be incorporated into the final MP structure and ELAN structure to form MP_DCN and ELAN_DCN structures to improve the YOLOv7 network. The DCNv2 structure includes two key structures, namely, a variable convolution and a variable pooling layer. By introducing an additional offset, the feature extraction process is concentrated in the effective information area, thereby improving the detector's ability to adapt to geometric changes in objects and ultimately improving its performance in detecting non-rigid objects (i.e., visible dust on construction sites).

[0012] After completing the construction of the corresponding target detection model, the image data corresponding to the current position to be tested will be obtained immediately, and then input into the target detection model of construction dust and prevention measures. The target detection model will output the corresponding target detection results of construction dust and prevention measures. This method greatly improves the accuracy and reliability of the output of the target detection results, and greatly helps the subsequent clients to have an accurate and reliable understanding of the construction dust and prevention measures.

[0013] Furthermore, the Backbone module implementation mechanism in S2 is as follows:

[0014] S201, the input image data is subjected to feature integration and extraction through four CBS structures with different convolution kernels and different step sizes to obtain a feature map. Each CBS structure includes a convolution layer (Conv2d), a normalization layer (BatchNorm) and a SILU activation function;

[0015] S202, input the obtained feature map into the ELAN module, the ELAN module includes multiple CBS structures; through the convolution of different branches, four feature maps are finally obtained, and the four feature maps are superimposed, and after superposition, the final feature extraction result is output after the [1,1] CBS structure;

[0016] S203, input the feature extraction result obtained in the ELAN module into the MPConv structure for downsampling processing to obtain a new feature map;

[0017] S204, inputting the new feature map obtained in S203 into the ELAN module, and outputting the corresponding first feature map C1;

[0018] S205, inputting the first feature map C1 into the MPConv structure first, outputting the corresponding first transit feature map, and then inputting the first transit feature map into the ELAN module, outputting the corresponding second feature map C2;

[0019] S206: Input the output second feature map C2 into the MP_DCN structure first, output the corresponding second transit feature map, and then input the second transit feature map into the ELAN_DCN structure, and output the corresponding third feature map C3.

[0020] Beneficial effects: In this solution, by setting up the MP_DCN structure and the ELAN_DCN structure, the accuracy of the corresponding feature extraction is higher, and the feature extraction of the data corresponding to the construction dust can be performed efficiently and accurately under the complex environment of the construction site, providing a reliable and fast basis for the subsequent detection of construction dust, reducing the occurrence of missed detection or false detection. At the same time, it has significant advantages over the existing technology in terms of average accuracy and detection speed of object detection.

[0021] Furthermore, the implementation mechanism of the Head module in S2 is as follows:

[0022] S211, according to the first feature map C1, the second feature map C2, and the third feature map C3 output by the Backbone module, the third feature map C3 is input into the SPPCSPC structure, and the corresponding feature maps T1 and T9 are output after processing; the SPPCSPC structure includes multiple CBS results and a maximum pooling layer;

[0023] S212, input the output feature map T1 into the CBS structure for convolution and upsampling, and after processing, concatenate it with the second feature map C2 to form a corresponding feature map T2;

[0024] S213, input the feature map T2 into the ELAN2 structure, and output the corresponding feature maps T3 and T4;

[0025] S214, input the output feature map T3 into the CBS structure, perform upsampling processing, concatenate it with the first feature map C1 after processing, and then input it into the ELAN2 structure to obtain corresponding feature maps T5 and T6;

[0026] S215, input the obtained feature map T5 and feature map T4 into MPConv2 for convolution processing, input the processed data into ELAN2 structure processing for further processing, and obtain the corresponding feature map T7 and feature map T8;

[0027] S216, the feature map T7 and the feature map T9 are input to MPConv2 for convolution processing, and the obtained feature map enters ELAN2 and RepConv structure to obtain the first final feature map P1;

[0028] The feature map T6 is input into the RepConv structure, and the second final feature map P2 is output;

[0029] The feature map T8 is input into the RepConv structure, and the third final feature map P3 is output;

[0030] S217, input the output first final feature map P1, the second final feature map P2, and the third final feature map P3 into the Detect structure for classification detection, and return the coordinate position.

[0031] Further, the S3 includes:

[0032] S30, dividing the construction site image data with annotated information into a training set, a validation set, and a test set in a ratio of 7:1:2;

[0033] S31, inputting the construction site image data corresponding to the training set into the target detection model of construction dust and prevention measures, and performing iterative training;

[0034] S32. After each iteration of training is completed, the trained construction dust and prevention measures target detection model is output. By inputting the construction site image data in the verification set, the training effect of the trained construction dust and prevention measures target detection model is verified to determine whether the training effect meets the standard; if not, the training is continued; if so, the training is stopped, and the construction dust and prevention measures target detection model is tested with the construction site image data in the test set to output the test results; based on the current test results and the previous test results, the feasibility of the corresponding construction dust and prevention measures target detection model is judged; if not feasible, the training is continued; otherwise, the corresponding construction dust and prevention measures target detection model is output.

[0035] Furthermore, the labeling information includes visible dust, face protection, no face protection, head protection, no head protection, spray vehicle, covering fabric, particle monitor and spray tank.

[0036] Furthermore, it also includes S5: after obtaining the corresponding target detection result, based on the target detection result, formulating a dust prevention and control measure plan corresponding to the current position to be measured, and sending it to the client.

[0037] Beneficial effects: In this scheme, not only the corresponding position of the construction dust corresponding to the current measured position is determined, but also the corresponding prevention and control measures are made, so as to realize accurate detection of construction dust on the construction site and carry out precise prevention and control, greatly reducing the existence time of construction dust, ensuring the safety of operators on the construction site, realizing multi-target detection and improving detection efficiency.

[0038] Further, judging whether the training effect meets the standard in S32 includes:

[0039] When the trained construction dust and prevention measures target detection model is output after each iteration of training, the loss function value is calculated through the Wise-IoU loss function, and based on the loss function value, the weight corresponding to the construction dust and prevention measures target detection model is updated through the gradient gain allocation strategy, and the predicted value and actual value corresponding to the trained and updated construction dust and prevention measures target detection model are calculated, and the predicted value and actual value corresponding to the construction site image data in a preset number of training sets are selected, and the prediction accuracy corresponding to each construction site image data is calculated. Based on the prediction accuracy, the corresponding correct proportion is calculated, and the correct proportion is compared with the preset proportion threshold. When the correct proportion is greater than or equal to the preset proportion threshold, it is judged that the training effect of the trained construction dust and prevention measures target detection model meets the standard, otherwise it is judged that the training effect of the trained construction dust and prevention measures target detection model does not meet the standard.

[0040] Beneficial effects: In this scheme, the Wise-IoU loss function is used to evaluate the training effect of the trained target detection model of construction dust and prevention measures, which greatly improves the corresponding evaluation accuracy and reliability.

[0041] Wise-IoU is used to replace IoU to evaluate the detection quality of the model, and the gradient gain distribution strategy is used to effectively attenuate the impact of low-quality samples on the loss function, aiming to balance the learning efficiency of low-quality samples and high-quality samples and improve the overall performance of the detector.

[0042] The present invention also provides a target detection and early warning system for construction dust and prevention measures, using the above-mentioned target detection and early warning method for construction dust and prevention measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a flow chart of a construction dust target detection and early warning method based on an improved YOLOV7 network in Embodiment 1 of the present invention;

[0044] Figure 2 This is a diagram of the improved YOLOV7 network structure in Embodiment 1 of the present invention;

[0045] Figure 3 It is a structural diagram corresponding to the MP_DCN structure and the ELAN_DCN structure in the first embodiment of the present invention;

[0046] Figure 4 This is a diagram showing the target detection results in Example 1 of the present invention.

[0047] Figure 5 This is a comparison diagram between different models in Example 1 of the present invention. DETAILED DESCRIPTION

[0048] The following is further described in detail through specific implementation methods:

[0049] Embodiment 1

[0050] A target detection and early warning method for construction dust and prevention measures, basically as follows Figure 1 As shown, the following steps are included:

[0051] S1. Obtain a construction site image dataset with annotated information from a historical database; in this embodiment, the annotated information includes visible dust, face protection, no face protection, head protection, no head protection, spray truck, covering fabric, particle monitor, and spray tank. In this embodiment, the construction site image dataset is formed by taking photos and sampling at the construction site, expanding the data, and downloading the public dataset on the Internet to obtain the relevant data of construction dust and prevention measures.

[0052] S2. Construct a target detection model for construction dust and prevention measures (YOLOv7-DCN-W). The target detection model for construction dust and prevention measures includes a Backbone module and a Head module. The corresponding improved YOLOV7 network structure is shown in the figure below: Figure 2 The Backbone module includes a CBS structure, an ELAN module, an MPConv structure, an MP_DCN structure, and an ELAN_DCN structure; the MP_DCN structure is formed by incorporating a DCDv2 structure into the final MP structure, and the ELAN_DCN structure is formed by incorporating a DCDv2 structure into the final ELAN structure;

[0053] The Backbone module implementation mechanism in S2 is as follows:

[0054] S201, the input image data is subjected to feature integration and extraction through four CBS structures with different convolution kernels and different step sizes to obtain a feature map. Each CBS structure includes a convolution layer (Conv2d), a normalization layer (BatchNorm) and a SILU activation function;

[0055] S202, input the obtained feature map into the ELAN module, the ELAN module includes multiple CBS structures; through the convolution of different branches, four feature maps are finally obtained, and the four feature maps are superimposed, and after superposition, the final feature extraction result is output after the [1,1] CBS structure;

[0056] S203, input the feature extraction result obtained in the ELAN module into the MPConv structure for downsampling processing to obtain a new feature map;

[0057] S204, inputting the new feature map obtained in S203 into the ELAN module, and outputting the corresponding first feature map C1;

[0058] S205, inputting the first feature map C1 into the MPConv structure first, outputting the corresponding first transit feature map, and then inputting the first transit feature map into the ELAN module, outputting the corresponding second feature map C2;

[0059] S206, input the output second feature map C2 into the MP_DCN structure first, output the corresponding second transit feature map, then input the second transit feature map into the ELAN_DCN structure, and output the corresponding third feature map C3. The structure diagram corresponding to the MP_DCN structure and the ELAN_DCN structure is shown in FIG. Figure 3 In this embodiment, as shown. Figure 5As shown, the target detection model for construction dust and prevention measures (YOLOv7-DCN-W) corresponding to this embodiment has significant advantages over other existing models, such as YOLOv7 and YOLOv7-DCN, in terms of the average progress and detection output of the corresponding object detection.

[0060] The implementation mechanism of the Head module in S2 is as follows:

[0061] S211, according to the first feature map C1, the second feature map C2, and the third feature map C3 output by the Backbone module, the third feature map C3 is input into the SPPCSPC structure, and the corresponding feature maps T1 and T9 are output after processing; the SPPCSPC structure includes multiple CBS results and a maximum pooling layer;

[0062] S212, input the output feature map T1 into the CBS structure for convolution and upsampling, and after processing, concatenate it with the second feature map C2 to form a corresponding feature map T2;

[0063] S213, input the feature map T2 into the ELAN2 structure, and output the corresponding feature maps T3 and T4;

[0064] S214, input the output feature map T3 into the CBS structure, perform upsampling processing, concatenate it with the first feature map C1 after processing, and then input it into the ELAN2 structure to obtain corresponding feature maps T5 and T6;

[0065] S215, input the obtained feature map T5 and feature map T4 into MPConv2 for convolution processing, input the processed data into ELAN2 structure processing for further processing, and obtain the corresponding feature map T7 and feature map T8;

[0066] S216, the feature map T7 and the feature map T9 are input to MPConv2 for convolution processing, and the obtained feature map enters ELAN2 and RepConv structure to obtain the first final feature map P1;

[0067] The feature map T6 is input into the RepConv structure, and the second final feature map P2 is output;

[0068] The feature map T8 is input into the RepConv structure, and the third final feature map P3 is output;

[0069] S217, input the output first final feature map P1, the second final feature map P2, and the third final feature map P3 into the Detect structure for classification detection, and return the coordinate position.

[0070] S3, transferring the construction site image dataset with labeled information to the target detection model of construction dust and prevention measures, and iteratively training the model;

[0071] The S3 includes:

[0072] S30, dividing the construction site image data with annotated information into a training set, a validation set, and a test set in a ratio of 7:1:2;

[0073] S31, inputting the construction site image data corresponding to the training set into the target detection model of construction dust and prevention measures, and performing iterative training;

[0074] S32. After each iteration of training is completed, the trained construction dust and prevention measures target detection model is output. By inputting the construction site image data in the verification set, the training effect of the trained construction dust and prevention measures target detection model is verified to determine whether the training effect meets the standard; if not, the training is continued; if so, the training is stopped, and the construction dust and prevention measures target detection model is tested with the construction site image data in the test set to output the test results; based on the current test results and the previous test results, the feasibility of the corresponding construction dust and prevention measures target detection model is judged; if not feasible, the training is continued; otherwise, the corresponding construction dust and prevention measures target detection model is output.

[0075] The step S32 of judging whether the training effect meets the standard includes:

[0076] When the trained construction dust and prevention measures target detection model is output after each iteration of training, the loss function value is calculated through the Wise-IoU loss function, and based on the loss function value, the weight corresponding to the construction dust and prevention measures target detection model is updated through the gradient gain allocation strategy, and the predicted value and actual value corresponding to the trained and updated construction dust and prevention measures target detection model are calculated, and the predicted value and actual value corresponding to the construction site image data in a preset number of training sets are selected, and the prediction accuracy corresponding to each construction site image data is calculated. Based on the prediction accuracy, the corresponding correct proportion is calculated, and the correct proportion is compared with the preset proportion threshold. When the correct proportion is greater than or equal to the preset proportion threshold, it is judged that the training effect of the trained construction dust and prevention measures target detection model meets the standard, otherwise it is judged that the training effect of the trained construction dust and prevention measures target detection model does not meet the standard. In this embodiment, for example, a batch of data such as 32 pictures is selected from this test set, and then the prediction accuracy corresponding to each picture is calculated. After that, based on the prediction accuracy of all pictures, the corresponding correct proportion is calculated, that is, the correct number and the total picture data of the batch are used to calculate the corresponding correct proportion, and whether the training effect of this model meets the standard is determined based on the correct proportion.

[0077] Wise-IoU is used to replace IoU to evaluate the quality of model detection, and the gradient gain allocation strategy is used to effectively attenuate the impact of low-quality samples on the loss function, aiming to balance the learning efficiency of low-quality samples and high-quality samples and improve the overall performance of the detector. Among them, the gradient gain allocation strategy: when encountering low-quality data, its IoU calculation value may be equal to 0, resulting in gradient loss. First, Wise-IoU uses outliers to evaluate the quality of the prediction box (outliers: the degree to which the prediction box deviates from the true box). If the outlier of a prediction box is very high (i.e., the degree of deviation from the true box is large), the gradient weight of the sample may be reduced to avoid it from having an adverse effect on the model update. On the contrary, samples with low outliers (i.e., prediction boxes that are closer to the true box) will obtain higher gradient weights, thereby further strengthening their role in model training. The gradient loss problem is effectively avoided. In this embodiment, the traditional IoU loss function evaluates the ratio of the intersection area to the union area between the prediction box and the true box (annotation information). The larger the IoU, the higher the overlap. However, when some poor data appears, IoU may be 0, resulting in gradient loss and uneven gradient descent. In order to solve this problem, Wise-IoU uses outliers instead of IoU to evaluate the quality of anchor boxes, and effectively weakens the harm of low-quality samples to the gradient through the gradient gain distribution strategy. The specific usage logic is to use the Wise-Iou loss function to calculate, back-propagate the loss function value, update the weights of the target detection model of construction dust and prevention measures, and continue the next round of training.

[0078] S4, access the construction site monitoring system, obtain the image data corresponding to the current position to be tested, and transmit the image data as input to the trained construction dust and prevention measures target detection model, output the corresponding construction dust and prevention measures target detection results corresponding to the position to be tested, and send them to the client. The display diagram of the target detection result is as follows Figure 4 shown.

[0079] S5. After obtaining the corresponding target detection result, a dust prevention and control measure plan corresponding to the current position to be tested is formulated based on the target detection result and sent to the client. In this embodiment, the prevention and control measures are detected at the construction site, that is, whether the corresponding site has been dust prevention and control, if yes, then another dust prevention and control measure plan is formulated, if not, then the corresponding dust prevention and control measure plan is formulated and sent to the client for corresponding prevention and control.

[0080] This embodiment also discloses a target detection and early warning system for construction dust and prevention measures, using the above-mentioned target detection and early warning method for construction dust and prevention measures. In this embodiment, the corresponding system is deployed on a drone, and the drone is used to take photos and samples of the construction site. When dust is detected, the alarm service module of the integrated monitoring system is connected to remind the management personnel through the module.

[0081] The above is only an embodiment of the present invention. The common sense such as the known specific structure and characteristics in the scheme is described too much here. The ordinary technicians in the relevant field know all the common technical knowledge in the technical field of the invention before the application date or priority date, can know all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, which will not affect the effect of the implementation of the present invention and the practicality of the patent. The protection scope required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A target detection and early warning method for construction dust and prevention measures, characterized by: The following steps are involved: S1, obtain the construction site image dataset with annotation information from the historical database; S2. Construct a target detection model for construction dust and prevention measures; the target detection model for construction dust and prevention measures includes a Backbone module and a Head module; the Backbone module includes a CBS structure, an ELAN module, an MPConv structure, an MP_DCN structure, and an ELAN_DCN structure; the MP_DCN structure is formed by incorporating a DCDv2 structure into a final MP structure, and the ELAN_DCN structure is formed by incorporating a DCDv2 structure into a final ELAN structure; S3, transferring the construction site image dataset with labeled information to the target detection model of construction dust and prevention measures, and iteratively training the model; S4. Access the on-site monitoring system of the construction site, obtain the image data corresponding to the current position to be measured, and transmit the image data as input to the trained target detection model of construction dust and prevention measures, output the target detection result of construction dust and prevention measures corresponding to the position to be measured, and send it to the client.

2. A target detection and early warning method for construction dust and prevention measures according to claim 1, characterized in that: The Backbone module implementation mechanism in S2 is as follows: S201, the input image data is subjected to feature integration and extraction through four CBS structures with different convolution kernels and different step sizes to obtain a feature map. Each CBS structure includes a convolution layer (Conv2d), a normalization layer (BatchNorm) and a SILU activation function; S202, input the obtained feature map into the ELAN module, the ELAN module includes multiple CBS structures; through the convolution of different branches, four feature maps are finally obtained, and the four feature maps are superimposed, and after superposition, the final feature extraction result is output after the [1,1] CBS structure; S203, input the feature extraction result obtained in the ELAN module into the MPConv structure for downsampling processing to obtain a new feature map; S204, inputting the new feature map obtained in S203 into the ELAN module, and outputting the corresponding first feature map C1; S205, inputting the first feature map C1 into the MPConv structure first, outputting the corresponding first transit feature map, and then inputting the first transit feature map into the ELAN module, outputting the corresponding second feature map C2; S206: Input the output second feature map C2 into the MP_DCN structure first, output the corresponding second transit feature map, and then input the second transit feature map into the ELAN_DCN structure, and output the corresponding third feature map C3.

3. A target detection and early warning method for construction dust and prevention measures according to claim 2, characterized in that: The implementation mechanism of the Head module in S2 is as follows: S211, according to the first feature map C1, the second feature map C2, and the third feature map C3 output by the Backbone module, the third feature map C3 is input into the SPPCSPC structure, and the corresponding feature maps T1 and T9 are output after processing; the SPPCSPC structure includes multiple CBS results and a maximum pooling layer; S212, input the output feature map T1 into the CBS structure for convolution and upsampling, and after processing, concatenate it with the second feature map C2 to form a corresponding feature map T2; S213, input the feature map T2 into the ELAN2 structure, and output the corresponding feature maps T3 and T4; S214, input the output feature map T3 into the CBS structure, perform upsampling processing, concatenate it with the first feature map C1 after processing, and then input it into the ELAN2 structure to obtain corresponding feature maps T5 and T6; S215, input the obtained feature map T5 and feature map T4 into MPConv2 for convolution processing, input the processed data into ELAN2 structure processing for further processing, and obtain the corresponding feature map T7 and feature map T8; S216, the feature map T7 and the feature map T9 are input to MPConv2 for convolution processing, and the obtained feature map enters ELAN2 and RepConv structure to obtain the first final feature map P1; The feature map T6 is input into the RepConv structure, and the second final feature map P2 is output; The feature map T8 is input into the RepConv structure, and the third final feature map P3 is output; S217, input the output first final feature map P1, the second final feature map P2, and the third final feature map P3 into the Detect structure for classification detection, and return the coordinate position.

4. A target detection and early warning method for construction dust and prevention measures according to claim 3, characterized in that: The S3 includes: S30, dividing the construction site image data with annotated information into a training set, a validation set, and a test set in a ratio of 7:1:2; S31, inputting the construction site image data corresponding to the training set into the target detection model of construction dust and prevention measures, and performing iterative training; S32. After each iteration of training is completed, the trained construction dust and prevention measures target detection model is output. By inputting the construction site image data in the verification set, the training effect of the trained construction dust and prevention measures target detection model is verified to determine whether the training effect meets the standard; if not, the training is continued; if so, the training is stopped, and the construction dust and prevention measures target detection model is tested with the construction site image data in the test set to output the test results; based on the current test results and the previous test results, the feasibility of the corresponding construction dust and prevention measures target detection model is judged; if not feasible, the training is continued; otherwise, the corresponding construction dust and prevention measures target detection model is output.

5. The target detection and early warning method for construction dust and prevention measures according to claim 4 is characterized by: The labeling information includes visible dust, face protection, no face protection, head protection, no head protection, spray vehicle, covering fabric, particle monitor and spray tank.

6. A target detection and early warning method for construction dust and prevention measures according to claim 5, characterized in that: The method further includes S5: after obtaining the corresponding target detection result, formulating a dust prevention and control measure plan corresponding to the current position to be tested based on the target detection result, and sending the plan to the client.

7. A target detection and early warning method for construction dust and prevention measures according to claim 6, characterized in that: The step S32 of judging whether the training effect meets the standard includes: When the trained construction dust and prevention measures target detection model is output after each iteration of training, the loss function value is calculated through the Wise-IoU loss function, and based on the loss function value, the weight corresponding to the construction dust and prevention measures target detection model is updated through the gradient gain allocation strategy, and the predicted value and actual value corresponding to the trained and updated construction dust and prevention measures target detection model are calculated, and the predicted value and actual value corresponding to the construction site image data in a preset number of training sets are selected, and the prediction accuracy corresponding to each construction site image data is calculated. Based on the prediction accuracy, the corresponding correct proportion is calculated, and the correct proportion is compared with the preset proportion threshold. When the correct proportion is greater than or equal to the preset proportion threshold, it is judged that the training effect of the trained construction dust and prevention measures target detection model meets the standard, otherwise it is judged that the training effect of the trained construction dust and prevention measures target detection model does not meet the standard.

8. A target detection and early warning system for construction dust and prevention measures, characterized in that: A target detection and early warning method for construction dust and prevention measures using any one of claims 1 to 7 above.

Images