Intelligent pest monitoring: computer vision for mealybug classification and quantification on sticky traps

This thesis explores how computer vision techniques can be used to improve the accuracy and reliability of identifying male mealybugs on sticky traps. The study focuses on classifying and quantifying three major mealybug species that affect the Mediterranean region: Delottococcus aberiae, Planococcus citri, and Pseudococcus longispinus. A total of 3,835 images were labeled, analyzed, and distributed as follows: D. aberiae (1,025; 26.72%), P. citri (1,823; 47.53%), and P. longispinus (987; 25.75%). The aim is to reduce the labor-intensive nature of manual identification and increase the precision of species classification. Notably, P. citri, previously the most challenging species for the model to classify, is now the most represented class in the dataset. The normalized confusion matrix reveals strong overall performance. All three species achieved high classification accuracy, with true positive rates around 93–94%. While some confusion persists with the background class, particularly for P. citri (41%) and D. aberiae (32%), the results demonstrate clear progress, especially in the detection of P. citri. These findings suggest that the model is becoming more balanced across species, addressing earlier limitations in classification. These results are an important first step toward creating a faster and more accurate system for monitoring mealybug populations. This method could help collect useful information about how pest populations change over time, support better predictions to guide decisions, and, in the long term, help develop better strategies to manage these pests in an effective and sustainable way.