YOLO-AF:结合注意力机制和焦点卷积的垃圾堆检测模型

YOLO-AF: A Garbage Heap Detection Model Combining Attention Mechanism and Focus Convolution

随着城镇化进程的不断加速,垃圾堆成为城市管理的重要问题,智能化的垃圾堆检测算法,可以有效提高环境管理的水平和效率。在YOLOv7检测算法的基础上,通过引入注意力机制、改进MPConv模块以及优化损失函数对模型进行改进,提出了YOLO-AF垃圾堆检测模型。引入注意力机制和使用焦点模块增强特征信息解决垃圾堆和背景相似导致检测精度低问题,综合利用二元交叉熵损失函数和Kullback-Leibler散度损失函数解决垃圾分类不平衡问题。实验结果表明,在多类型垃圾堆检测任务中,YOLO-AF垃圾堆检测模型相较于YOLOv7表现出更高的精度、召回率和mAP,性能得到显著提升,训练300轮次的模型mAP值为92.45%。在同等条件下,检测效果优于当前主流的目标检测算法,可以较好地满足垃圾堆实时工业检测的需求。

With the continuous acceleration of urbanization, garbage heaps have become an important problem in urban management, and intelligent garbage heap detection algorithms can effectively im-prove the level and efficiency of environmental management. On the basis of the YOLOv7 detection algorithm, the YOLO-AF garbage heap detection model was proposed by introducing the attention mechanism, improving the MPConv module and optimizing the loss function. The attention mechanism is introduced and the focus module is used to enhance the feature information to solve the problem of low detection accuracy caused by the similarity of garbage and background. The binary cross entropy loss function and Kullback-Leibler divergence loss function are comprehensively used to solve the unbalanced problem of garbage classification. Experimental results show that the YOLO-AF garbage heap detection model shows higher accuracy, recall rate and mAP than YOLOv7 in multi-type garbage heap detection tasks, and the performance is significantly improved, and the mAP value of the model after 300 rounds of training is 92.45%. Under the same conditions, the de-tection effect is better than the current mainstream object detection algorithm, which can better meet the needs of real-time industrial detection of garbage heaps.