Weighted Deformable Convolution Network with IOU-boundary loss for Solid Waste Detection

Abstract

It is of great significance for the detection of solid waste for environmental protection. In recent years, object detection methods based on deep learning have been widely studied. Different from regular objects such as airplanes or buildings, solid waste commonly has arbitrary shapes and the boundaries are always hard to distinguish. In this paper, a weighted deformable convolution module was proposed in which the offset and weight for each sampling location of the feature map were further considered. In this way, the feature representation of irregular objects can be enhanced and the regions can be extracted effectively. Furthermore, a combined IOU-boundary loss, including the Intersection-over-Union (IOU) and the aspect ratio (AR) losses, was designed to regress these irregular solid waste. Finally, a solid waste dataset was constructed manually to evaluate the proposed method, and it shows that the proposed method outperforms other traditional object detection methods, such as FPN, YOLO and CenterNet.