The detection of individual silkworms is an essential subtask for the detection of diseased silkworms by combining visual and trajectory characteristics. Due to the high-density and compact conditions, the detection model is required to have high efficiency and reliable accuracy. However, most existing networks fail to provide a trade-off between accuracy and efficiency. To alleviate this problem, a feature enhancement You Only Look Once (FeYOLO) is proposed by integrating two feature enhancement modules into the YOLOv7-tiny network. Specifically, a channel recalibration model (CRM) is designed based on the channel attention mechanism to learn more diversified expressions and to recalibrate feature channels, for which a soft-feature aggregation is developed to preserve the intrinsic relationship of features using a linear transformation layer. Then, a spatial enhancement module (SEM) is proposed to enhance detection capabilities by exploiting deep contextual information and applying dilated convolution at different dilated rates to expand the receptive fields. The head and tail of silkworms are regarded as detection targets to solve the problem of occluded and incomplete silkworms. Experimental results demonstrate that FeYOLO achieves 90.61 % mAP, 16.02 GFLOPs, and 87.34 FPS. Compared with the original YOLOv7 (YOLOv7l, YOLOv7x, and YOLOv7-tiny), state-of-the-art networks (RetinaNet, EfficientDet, YOLOv8s, YOLOv9s, and YOLOv10s), lightweight backbone-based models (MobileNetv1, MobileNetv2, MobileNetv3, and GhostNetv1), and image attention-based architectures (SENet, ECANet, CBAM, CANet, SimAM, and GCNet), FeYOLO achieves a notable improvement in balancing accuracy and efficiency for silkworm detection under complex conditions. This study not only offers valuable insights into silkworm disease identification by integrating visual and trajectory characteristics but also contributes to the advancement of object detection techniques in agriculture and biological applications.