DIG-YOTR: An End-to-End Environmental Pollution Detection Network for UAV Images

In response to the difficulty of monitoring and managing environmental pollution in rural areas, this paper designs an end-to-end algorithm for detecting environmental pollution in aerial images aimed at unmanned aerial vehicle systems, addressing the current challenges in monitoring rural environmental pollution. Firstly, to address the issue of convolutional neural networks' insufficient global receptive field for images, a novel Transformer structure detection head is designed to enhance the utilization of global features in images and realize an end-to-end network structure. Moreover, in order to enhance the detection of the model for small objects, Inner-IOU loss function is used to optimize the bounding box loss function and accelerate the convergence of the model. Finally, to improve the real-time performance of the detection algorithm, GSConv is utilized to optimize the convolution structure of the model, reducing computational complexity. Experimental results show that the model proposed in this paper achieves an average detection accuracy of 72.7% on a self-built environmental pollution dataset, with real-time performance reaching 120FPS, fully demonstrating the superiority of our algorithm in the field of environmental pollution detection.