一种改进的轻量级垃圾目标检测算法

垃圾分类问题的解决方法目前主要依靠垃圾处理厂人工分拣,其工作环境较差且自动化程度不高.为了提高垃圾分拣的速度与精度,以及为自动垃圾分拣设备提供算法解决参考方案,文章提出一种面向低功耗设备的轻量级垃圾目标检测算法Ghost-YOLO,该算法在保证轻量化的同时具有较高的垃圾检测精度.Ghost-YOLO算法是基于YOLOv...

An Improved Lightweight Garbage Target Detection Algorithm

At present, the solution to the problem of garbage classification mainly relies on manual sorting of garbage treatment plant, which has poor working environment and low degree of automation. In order to improve the speed and accuracy of garbage sorting, and to provide an algorithm solution reference scheme for automatic garbage sorting equipment, we propose a lightweight garbage target detection algorithm Ghost-YOLO for low-power devices, which has high precision of garbage detection while ensuring lightweight. The Ghost-YOLO algorithm is based on the YOLOv3 algorithm through a series of lightweight improvements. Firstly, feature extraction is performed on the input image by introducing the feature extraction network of the Ghost bottleneck lightweight module. Secondly, through the improved lightweight feature fusion layer, the down-sampling link is added, and the features are fused twice, so that the network has a stronger ability to detect small objects and the position of the regression box is more accurate. Experiments show that the amount of parameters of the Ghost-YOLO algorithm model is reduced by 95.96% compared to the original YOLOv3,which greatly reduces the amount of calculation and network parameters, and the overall algorithm is compressed to 9.5 MB. The mAP of the garbage data set can reach 89.02%.