基于改进SOLOv2的家蚕图像实例分割的应用与研究

Application and Research of Silkworm Image Instance Segmentation Based on Improved SOLOv2

在家蚕养殖过程中,通常需要人工检查家蚕是否发生病变,但人工检查存在效率低、检查不全面等问题。为解决这些问题,本研究提出一种新颖的家蚕图像实例分割算法(多级残差特征融合Multi-ResFF-SOLOv2),该算法引入多级残差模块,分别是梯度、傅立叶和空间通道注意力残差模块,从而提取出更高维的语义信息。实验结果分析表明,该算法的平均精确度(Mean Average Precision,mAP)为96.7%,模型经过量化后,其推理速度为61fps。结果为家蚕农业养殖的智能化发展提供了有效的解决方案。

In silkworm farming,manual inspection for diseases is inefficient and incomplete.To address these issues,this study proposes a novel silkworm image instance segmentation algorithm called Multi-ResFF-SOLOv2.The algorithm incorporates multi-level residual modules,including gradient,Fourier,and spatial channel attention residual modules,to extract higher-dimensional semantic information.Experimental results show that the algorithm achieves a mean average precision(mAP)of 96.7%and,after model quantization,an inference speed of 61fps.This provides an effective solution for the intelligent development of silkworm agriculture.