In order to address the problems of small target scale, high overlap, and complex background of cotton leaves affected by aphids in field environments, traditional detection methods often struggle to balance accuracy and real-time performance, resulting in limited effectiveness in practical applications. Therefore, an improved lightweight detection model was proposed. On the basis of YOLO v8n, an adaptive downsampling module ( ADown) was added to enhance local feature extraction, a spatially enhanced attention module (SEAM) was introduced to achieve multi-scale information exchange, and a Focal - EloU loss function was constructed for dense small object perception, effectively improving the model's adaptability to small objects and complex backgrounds. The experimental results on the self built dataset showed that the improved model achieved significant improvement in detection accuracy while significantly reducing the number of parameters. The average accuracy reached 96.2%. In addition, the ablation study further demonstrated the effectiveness of each module in enhancing the overall performance of the cotton aphid infestation severity detection model. The comprehensive results indicated that the improved model balanced detection accuracy and model deployment efficiency while ensuring lightweight, providing a feasible solution for intelligent monitoring of agricultural pests.