Remote Sensing Identification And Dynamic Change Detection Of Anthropogenic Disturbed Areas

Due to the current rapid socio-economic development,soil erosion caused by human production and construction activities is becoming increasingly serious,featuring short duration,high frequency of occurrence and great damage to the surrounding environment.Therefore,it is urgent to monitor anthropogenic soil erosion parcels.At present,there are few studies on the identification,extraction and dynamic changes of large-scale anthropogenic soil erosion parcels based on remote sensing images,and the traditional methods of on-site investigation and visual interpretation are time-consuming,tedious and costly.With the advancement of remote sensing technology in China,the temporal and spatial resolution of remote sensing data is constantly improving.,while the intricate nature of remote sensing images coupled with the diverse range of features present a challenging obstacle in the detection and surveillance of the dynamic alterations occurring in anthropogenic soil erosion parcels.To tackle this issue,the paper conducts a thorough investigation and enhancement of the U-shaped network,and introduces a novel semantic segmentation framework(referred to as IUNet++)built on top of the improved UNet++ model。This study selects 21 counties in Hubei Province in 2021 and 5 district in Wuhan in 2020 and 2021 as the sample area for the study.The main work is as follows:(1)In order to obtain a large receptive field while reducing information loss and optimizing boundary segmentation accuracy,the Boundary Constraints and Jagged Hybrid Dilated Convolution Channel Chuffling Module(BCJHDC)is constructed,Which not only overcomes the issue of local information loss resulting from dilated convolution,cavity convolution,but also improves the segmentation accuracy and edge effect of the model for different scales of anthropogenic soil erosion parcels.Compared with the original UNet++ model,the ablation experiments show that the addition of this module demonstrates an enhancement of 3.38% and2.20% in the Intersection over Union(IOU)and F1-score metrics,respectively.(2)To further improve the accuracy of recognition,the Polarized Self-attention Module(PSA)is added to the ecoding layer to learn semantic information at the pixel level to mine finer,higher quality features.It has higher resolution in channel and spatial dimensions than other attention mechanisms,it could reduce feature loss.Compared to the original UNet++model,the ablation experiments show that the addition of this module has a 1.98% and 1.31%improvement in Intersection over Union(IOU)and F1-score metrics,respectively.(3)In order to achieve superior recognition accuracy,the IUNet++ model is developed by merging the aforementioned modules and by employing residual convolution in place of standard convolution in the encoder to prevent gradient disappearance and elevate feature extraction efficiency.This enables the model to effectively extract anthropogenic soil erosion parcels.The model has the highest IOU and F1-score values compared with the seven currently mainstream models,indicating the effectiveness of the model improvement.And based on this model,the anthropogenic soil erosion parcels of the remaining 73 counties in Hubei Province in 2021 were identified with the existing data(9 counties without image data),and a distribution map of anthropogenic soil erosion parcels in Hubei Province was constructed.(4)The dynamic change detection of anthropogenic soil erosion parcels was converted into a multi-classification problem,and the two-phase remote sensing images in Wuhan in 2020 and 2021 were synthesized into six-band data,combined with IUNet++ for classification,and the change areas and categories of anthropogenic soil erosion parcels were obtained,compared with the original UNet++,the Mean Intersection over Union(MIOU)was increased by 3.78%.The Chinese soil erosion model(The Chinese Soil Loss Equation(CLSE)model calculated the soil erosion grade in Wuhan in 2020 and 2021,and analyzed the soil erosion changes based on the distribution of anthropogenic soil erosion parcels.