Detection of leaf miner in sweet potato crops through image analysis using machine learning-based models.

The leaf miner (Liriomyza huidobrensis) represents a critical threat to sweet potato production, causing irreversible damage that traditional visual inspection fails to mitigate efficiently due to its subjectivity and slowness. The objective of the present study was to automate the detection of this pest through image analysis using deep learning models. Methodologically, the dataset "camote_minador" was constructed and annotated, comprising 751 images collected from fields in Lambayeque, Peru, to which dynamic data augmentation techniques were applied to ensure training variability. The performance of the YOLOv8s and YOLOv11s architectures was comparatively evaluated under standardized hyperparameter configurations. The results demonstrated the technical superiority of the YOLOv11s model, which achieved a Precision of 73.20%, a Recall of 66.72%, and a mAP@50 of 71.63%, outperforming its predecessor and evidencing a greater ability to discriminate between pest galleries and background noise. Furthermore, the operational feasibility of a mobile prototype based on Tensor Flow Lite for mid-range devices was established. It is concluded that the implementation of optimized architectures such as YOLOv11s constitutes an effective, accessible, and scalable technological solution to strengthen phytosanitary monitoring in the agricultural sector.