Smart insect monitoring based on YOLOV5 case study: Mediterranean fruit fly Ceratitis capitata and Peach fruit fly Bactrocera zonata

The agricultural sector in Egypt is adversely affected by factors such as inadequate soil fertility and environmental hazards such as pestilence and diseases. The implementation of early pest prediction techniques has the potential to enhance agricultural yield. Bactrocera zonata and Ceratitis capitata, known as peach fruit fly and Mediterranean fruit fly, respectively, are the predominant pests that cause significant damage to fruits on a global scale. The present study proposes a deep learning-based approach for the detection and quantification of pests. The proposed approach entails the retrieval of data pertaining to the adhesive trap condition, followed by its examination and presentation through a mobile application. The YOLOV5 model has been implemented for the purpose of pest classification, localization, and quantification. In order to address the issue of a restricted dataset, a hybrid technique of transfer learning and data augmentation (copy and paste) was employed. The proposed approach offers an intelligent real time pest detection, thereby facilitating the prediction of treatment options. An application for smartphones has been developed to aid farmers and agricultural professionals in the management and treatment of pests. The proposed approach has the potential to aid farmers in identifying the existence of pests, thereby diminishing the duration and resources needed for farm inspection. As per the results of the conducted experiments, the proposed approach demonstrates a noteworthy increase in performance. The weighted average accuracy reaches 84%, while precision (P), mean average precision (mAP), and F1-score show enhancements of up to 15%, 18%, and 7% respectively.

Automated summary

A deep learning-based approach is proposed for the detection and quantification of pests in agriculture. The approach involves the retrieval of data using adhesive traps, followed by examination and presentation through a mobile application. A hybrid technique of transfer learning and data augmentation is employed to address the issue of a restricted dataset. The proposed approach offers real-time intelligent pest detection and enables prediction of treatment options.