Objective An algorithm was developed based on YOLOv8s to accurately identify the pests like Trichophysetis cretacea and thrips on jasmine plants.

Method Pest images were collected from field sticky traps in jasmine fields to construct the dataset. Taking YOLOv8s as the baseline model, a P2 layer for detecting small objects, coordinate attention (CA), and multi-scale feature fusion (MSFF) modules were introduced. mAP_0.5, precision and recall were adopted as evaluation metrics. Ablation experiments and comparative tests with mainstream object detection algorithms were carried out to verify the effectiveness of the improved modules and the overall performance of the model.

Result Ablation experiments demonstrated that the P2 detection layer, the CA module and the MSFF module progressively improved detection performance. Compared with the original YOLOv8s, the improved model achieved a gain of 2.94 percentage points in mAP_0.5; the precision of T. cretacea and thrips increased by 1.97 and 4.38 percentage points, and the recall increased by 0.19 and 0.35 percentage points, respectively. Compared with mainstream models such as YOLOv9t, the modified YOLOv8s achieved a good balance between detection accuracy and the number of parameters.

Conclusion The improved YOLOv8s model effectively enhances the detection precision and recall of T. cretacea and thrips in complex jasmine field environments while maintaining a low parameter count and computational cost, making it readily deployable on smart monitoring devices and providing a reference for the intelligent monitoring of jasmine pests.