<p>Silkworm farming is a vital specialty agricultural industry in China. As the core raw material for high-end textiles, silk is directly linked to the livelihoods of tens of millions of workers. However, with the continuous expansion of farming scales, the traditional manual management approach for silkworm instar identification and disease detection has revealed increasing limitations. Early-stage diseases in silkworms are highly concealed; if not detected and intervened in a timely manner, the widespread outbreak of diseases can lead to devastating economic losses. To address these challenges, this paper proposes an improved detection model based on YOLOv8, named YOLOv8s-CDM. The model introduces a 160 × 160 small-object detection branch, enabling the network to effectively capture tiny silkworm targets larger than 4 × 4 pixels, thereby enhancing small-object detection capability. It further integrates a Convolutional Block Attention Module (CBAM) to strengthen feature extraction and feature selection, and adopts the MPDIoU loss function to refine bounding box regression accuracy. Comparative experimental results demonstrate that, compared with the original YOLOv8s, the proposed model achieves a 2.4% increase in precision, a 0.9% improvement in recall, and a 1.7% rise in mean average precision (mAP) to 91.4%. Although the dataset used in this study covers multiple silkworm instars and representative disease samples, the collection scenarios are relatively limited. Future work will involve validation across more silkworm rearing bases, different seasons, and a broader range of disease conditions to further assess the model’s generalization capability.</p>