The insect detection and counting tool, a major focus in recent development, employs cutting-edge AI models for enhanced performance. It consists of a three-model pipeline, beginning with X-Decoder for segmenting trap areas in images. This is followed by insect detection using either YOLOv8 for specific insect classes or GroundingDINO for open-vocabulary detection. The final step involves SAM-HQ, which extracts individual insect segmentation masks and counts them.

Validation results are promising, with an F1 score of 0.903 using GroundingDINO, indicating low rates of false positives and negatives. This tool efficiently monitors pest evolution, as demonstrated by tracking insect increases in traps over time. For scenarios requiring detection of specific insect classes, YOLOv8 is used. This necessitates a comprehensive, labeled dataset, like the one re-published on the Roboflow platform, covering various insect classes crucial for agricultural and ecological studies.

The tool's performance, validated by robust F1 and mAP scores, confirms its accuracy in classifying and counting insects. This advancement marks significant progress in pest monitoring technology, combining state-of-the-art AI models with traditional techniques to provide a nuanced and effective approach for various environmental conditions and insect types.