Research On Machine Identification And Change Detection Method Of Construction Waste Based On UAV Remote Sensing

Accurate identification and change detection of construction waste is the key to solving the problem of urban environmental pollution and construction waste siege.The construction waste piled up in the open has adverse effects on the urban environment and residents’health.The lack of timely and accurate dynamic monitoring data of construction waste piled up has undoubtedly increased the difficulty of supervision by the city management department.In view of the current problems of rapid identification and accurate positioning of construction waste,research methods for construction waste identification,classification and change detection suitable for UAV remote sensing images,and establish a high-accuracy,fast and practical semantic segmentation and change detection model.New technologies and new approaches for engineering application of construction waste identification and extraction based on human-machine remote sensing provide scientific reference for pilot urban planning and construction waste management.The research content and results of this paper are as follows:（1）Making a dataset for semantic segmentation and change detection of construction wasteTaking Pingdingshan City,Henan Province,and Jining City,Shandong Province as the research areas,the DJI Phantom 4 RTK multi-rotor drone was used to obtain orthophoto aerial images of construction waste storage areas.After a series of processing techniques such as image screening,image stitching and orthorectification,a semantic segmentation dataset with a sample size of 25,620 is formed.Histogram matching and georeferencing were performed on the image data of the same storage site of construction waste in different periods,and a total of573 pairs of change detection datasets with before and after time series were generated.（2）Construction waste machine identification method based on multi-semantic segmentationUsing the above-mentioned construction waste semantic segmentation dataset,based on the Pytorch deep learning environment,the PSPNet,Deeplab V3+and U-Net network models are used to build recognition models and carry out construction waste recognition experiments.After training and verification,the recognition results of the verification samples are counted one by one.The results show that the overall performance and recognition accuracy of the U-Net network model is the highest,the average intersection ratio is 91.64%,a pixel accuracy value of 95.43%,and the₁score is 0.955.On the basis of the above identification experiments,more detailed construction waste classification experiments were carried out.First,the model is optimized by increasing the depth of the original U-Net network to obtain the IUNet network model.Secondly,based on image processing technology,three image features of edge,texture and color are extracted from the UAV image,and these features are input into the IUNet model in different combinations to build a construction waste IUNet-IF classification model covering the dust net.The results show that the combination of texture features,color features and the original image is the optimal classification model,the pixel accuracy index is as high as 98.96%,the average intersection ratio is as high as 93.60%,and the class average pixel accuracy is as high as 97.18%.（3）Binary area change detection method for construction wasteBased on the U-Net network,a spatial attention module is introduced into the skip connection part to construct the A-UNet construction waste change detection model.The image channels of two different periods are combined to form a six-channel data,which is input into the improved change detection model for training.The results show that the recall rate of the model is as high as 93.1%,and the change detection accuracy is as high as 91.78%.The introduced spatial attention module can effectively improve the robustness of the model.To sum up,this study aims at the remote sensing data of UAVs,combined with deep learning and image processing methods to construct the IUNet-IF construction waste classification model,and introduces the spatial attention module based on the U-Net network to construct the A-UNet construction waste change detection model.It realizes the identification,classification and change detection of construction waste,and forms a complete machine identification and dynamic monitoring system for construction waste.In practical application,the system can quickly and accurately identify the construction waste covering the dust net,obtain its reliable type,and detect its changing area.The model runs well and achieves the expected results,which can provide strong support for the engineering application of remote sensing UAV remote sensing monitoring of construction waste.