基于YOLOv6的青藏高原畜种图像的识别研究

Research on the Recognition of Livestock Images in the Qinghai-Tibet Plateau Based on YOLOv6

针对遮挡条件下的青藏高原畜种图像识别错误率高、漏检率高的问题,本文提出BC-YOLOv6算法,将YOLOv6的Backbone与双向结构的Transformer层的BiFormer相结合,同时在其架构中引入坐标注意力机制模块(CA),提高对小目标的识别率,改进非极大值抑制(NMS),设计相关函数,提高召回率。实验结果表明,BC-YOLOv6比原模型的精度、召回率、平均精度均值分别提高了16.6%、8.9%、21.9%,有效解决了遮挡问题。

Aiming at the problem of high error rate and missed detection rate in animal breed image recognition on the Qinghai-Tibet Plateau under occlusion conditions we propose the BC-YOLOv6 algorithm,which combines the backbone of YOLOv6 with the BiFormer of the bidirectional Transformer layer.At the same time,the Coordinate Attention for Effective Mobile Network Design(CA)module in its architecture is introduced to improve the recognition rate of small targets,NMS,design relevant functions,and increase recall rate.The experimental results show that BC-YOLOv6 has improved the accuracy,recall,and average accuracy of the original model by 16.6%,8.9%,and 21.9%,respectively.