Detection of Whiteflies in Plants using Deep Learning

Whiteflies, One of the devastating pests seen in high-yielding crops that cause mass damage to agriculture due to their transmitting capability through a broad range of scopes. These pest species cause insufficient agricultural production on a large scale yields, so recognizing these insects in a larger area field is difficult and time-consuming for human observation. To overcome such a situation, in this paper, we used an object detection model known as Yolov5 with a nano variant, which helps detect thousands of small whiteflies rapidly with fewer parameters and power efficiency. Besides conventional pre-processing techniques, tilling images at inference helped to detect tiny whiteflies effectively and classified the abundance level based on counts. The model is fine-tuned with the pre-trained weight, and the result is evaluated using the Cassava Whitefly Dataset. Using the proposed methodology, the model achieved an average precision of 0.95@0.5 threshold with 0.92 F1-score. The experimental result demonstrates that the single-stage detector performs finer than the two-stage detector on speed and accuracy using tilling. The result of the proposed system increases the F1-score and average precision by 22% and 20% from the Raw data.