Automatic Detection and Counting of Tuta Absoluta Insect in Trap Images
Abstract
The integration of artificial intelligence (AI) into agriculture offers solutions to challenges such as pest control. AI can improve productivity and sustainability through precision agriculture. This paper presents an automatic system for identifying and counting Tuta absoluta pests in trap images, integrated into a monitoring platform. The platform uses a biological defensive system for sustained pest control. The solution employs ImageAI deep learning algorithms to detect and classify pests, using YOLOv3 and TinyYOLOv3 models. We provide assessments of the performance and resource consumption of the evaluated models. YOLOv3 achieved a detection precision of 95.28% in images with 10-50 insects, decreasing to 87.51% for around 100 insects. Despite YOLOv3 demonstrating higher precision in the detection of the number of insects, the Tiny YOLOv3 model was shown to be 4.5 times faster in the training process and occupies almost 8 times less storage space.
