【摘要】 水面漂浮垃圾不断增多引起关注，针对水面漂浮垃圾边缘信息模糊的问题，提出E-MP模块，在MPConv的基础上添加Laplacians,Sobel-dx和Sobel-dy增强小目标水面漂浮垃圾的边缘信息。针对小目标漂浮垃圾仅占据图像少量像素的现象，引入了Biformer注意力模块。Biformer利用前后两个方向的上下文信息，更好地捕捉序列中的依赖关系，同时降低背景信息对检测目标物体带来的一部分影响。在此基础上引入SIoU来构建损失函数，将边界区域作为目标区域来进行加权，可以更好地捕捉目标的边界信息，从而提高检测精度。在Flow-Img子数据集上进行了大量实验，实验结果表明，YOLOv7-edge模型比原来的模型检测精度更高，mAP@0.5和mAP@0.5:0.95分别提高了7个百分点和5个百分点。更多还原

【Abstract】 The increasing amount of floating garbage on the water surface has attracted people’s attention. In response to the problem of blurred edge information of floating garbage on the water surface, this article proposes an E-MP module,which adds Laplacians, Sobel-dx, and Sobel-dy to enhance the edge information of small floating garbage targets on the water surface based on MPConv. In response to the phenomenon that small floating garbage targets only occupy a small number of pixels in the image, a Biformer attention module has been introduced. The Biformer utilizes contextual information from both the front and back directions to better capture dependencies in the sequence while reducing some of the impact of background information on detecting the target object. On this basis, introducing SIoU to construct a loss function and weighting the boundary region as the target region can better capture the boundary information of the target and improve detection accuracy. A large number of experiments are conducted on the Flow-Img sub dataset, which shows that the YOLOv7 edge model had higher detection accuracy than the original model, mAP@0.5 and mAP@0.5 0.95 increased by 7 % and 5 % respectively.更多还原