一种改进型YOLOv5草莓检测算法

An Improved YOLOv5 Strawberry Detection Algorithm

针对现有的草莓检测算法模型参数量大、准确率低、实时性差等问题,提出一种改进型YOLOv5草莓检测算法。算法基于YOLOv5模型,骨干网络引入GhostConv和C3Ghost模块进行参数量压缩,构造轻量化模型;加入Cutout增强数据,增加训练样本的多样性,进而提高模型的泛化能力和抗干扰能力;通过引入Gather-Excite和Transformer注意力机制加强对草莓图像重要特征的关注,从而提升检测算法在复杂环境下的识别能力。实验显示,所提算法的平均精度均值1和平均精度均值2分别为97.7%和83.5%,参数量缩减为4.01 M,推理时间为26.3 ms。改进后的算法相比原算法具有识别速度快、定位准度高以及占用内存少的优势,在满足精准采摘工作要求的前提下可以提高采摘效率。

Aiming at the problems of large number of parameters,low accuracy and poor real-time performance of the existing strawberry detection algorithm model,an improved YOLOv5-based strawberry classification and detection algorithm was proposed.The algorithm was based on the YOLOv5s model.GhostConv and C3Ghost modules was introduced in the backbone network to compress the parameters and construct a lightweight model.The data was enhanced by Cutout so as to increase the diversity of training samples and improve the generalization ability and anti-interference ability of the model.By introducing the Gather-Excite and Transformer attention mechanism,the attention of the important features of strawberry images was strengthened so that the recognition ability of the detection algorithm in complex environments was improved.Experimental results showed that the mAP@0.5 and mAP@0.5:0.95 of the algorithm were 97.7%and 83.5%respectively,the parameter quantity was reduced to 4.01 M,and the inference time was 26.3 ms.Compared with the original algorithm,the improved algorithm has fast recognition speed,high positioning accuracy and less memory occupation.