Detection of Tuta absoluta larvae and their damages in tomatoes with deep learning-based algorithm

Abstract

Plant pests cause significant economic losses to the agricultural sector. Tuta absoluta is one of the most important pests of the tomato plant, which has a high financial return. Accurate and rapid identification of tomato plant pests is essential to increase sustainable agricultural productivity. Computer vision and machine learning techniques such as deep learning and especially Convolutional Neural Networks (CNN) have effectively identified various plant pests. Training CNN-based algorithms that allow us to classify and identify objects can enable faster and more accurate pest detection than human observation. We used YOLOv5 (You Look Only Once), a CNN-based object detection algorithm. One thousand two hundred photos of tomato leaves infested by the T. absoluta pest were edited to train the YOLOv5 algorithm. Images include T. absoluta larvae and galleries created in leaves by these larvae. Experimental results showed that using the YOLOv5 algorithm could categorize tomato plant leaves correctly and detect T. absoluta larvae, galleries with 80% and 70-90% accuracy (mAP), respectively. The research is promising that deep learning-based object identification algorithms can be significantly effective in detecting agricultural pests early and preventing excessive use of pesticides.