【摘要】 为了在资源有限的移动设备上对猪体质量进行准确和实时估计，提出了融合注意力机制的轻量级猪体质量实时估计方法，该方法分为分割阶段和体质量估计阶段。在分割阶段，为了能更准确地获得猪躯干分割图像，提出了边缘引导的轻量级分割网络EG-YOLO。在体质量估计阶段，为了更好地融合双视角特征用于体质量估计，基于改进的轻量级MobileNeXt子网络，构建了双视角特征融合的轻量级体质量估计网络，并在自建的猪体质量估计数据集上进行实验。结果表明，体质量估计的误差仅为3.99%。更多还原

【Abstract】 In order to achieve accurate and real-time pig weight estimation on resource-constrained mobile devices, a lightweight pig weight real-time estimation methocls was proposed, which consisted of a segmentation stage and a weight estimation stage. In the segmentation stage, an edge-guided lightweight segmentation network, named EG-YOLO, was presented to obtain more accurate segmentation images of the pig body. In the weight estimation stage, to better fuse dual-view features for weight estimation, a lightweight weight estimation network with dual-view feature fusion was constructed based on an improved lightweight MobileNeXt sub-network. Experimental results on a self-built pig weight estimation dataset show that the weight estimation error is only 3.99%.更多还原