Enhancing Silkworm Breeding Through Advanced Image Segmentation: A Comparative Study of U-Net, Mask R-CNN, And FCN

Accurate segmentation of male and female silkworms is a crucial task in the sericulture industry, where efficient sex differentiation contributes to optimized breeding and enhanced productivity. Traditionally, this task has been performed manually, which is labour-intensive and error-prone. To address these challenges, this paper presents a comprehensive approach to automated segmentation of silkworm images using advanced image processing techniques. Three different deep learning-based segmentation methods are explored: U-Net, Mask R-CNN, and Fully Convolutional Networks (FCNs). The U-Net model, known for its robust performance in biomedical image segmentation, is employed to delineate the contours of silkworms, ensuring precise boundary detection. Mask R-CNN, a region-based convolutional neural network, is used for instance segmentation, handling multiple silkworms within a single image and providing pixel-wise masks for each instance. Lastly, Fully Convolutional Networks (FCNs) offer a pixel-to-pixel mapping for semantic segmentation, assigning each pixel in the image a category related to the silkworm body or background. Results from these three methods are compared using metrics such as Intersection over Union (IoU) and pixel-wise accuracy, with U-Net achieving the highest IoU of 0.89, closely followed by Mask R-CNN with 0.87, and FCN with 0.85. These findings demonstrate that deep learning techniques can provide effective and reliable solutions for silkworm segmentation, significantly reducing manual effort and improving efficiency in large-scale sericulture operations.