Detection of Diatraea Saccharalis in images using convolutional neural networks

Agricultural pests are organisms capable of significantly impacting the yield and quality of cultivated crops. Traditionally, population control of insect pests has relied on methods such as trapping and subsequent analysis of captured individuals to implement specific control actions, such as the use of insecticides. However, advancements in Computer Vision and Deep Learning techniques offer promising ways for more efficient pest detection and management. This study aims to apply Convolutional Neural Networks (CNNs) to detect the insect pest Diatraea saccharalis, a major pest of sugarcane crops. A dataset comprising 945 training images and 470 test images of deceased insects collected from traps was compiled in order to train and test the model. The Yolov8 Computer Vision framework was employed for software implementation. Results indicate promising outcomes, with the trained CNN achieving 96.2% precision and 95.8% recall. The application of Computer Vision in pest management could lead to more timely and accurate detection of pests, reducing the need for widespread insecticide use, enabling specific interventions, and minimizing labor-intensive monitoring tasks. This research highlights the potential of Deep Learning methodologies to enhance agricultural pest management strategies by improving early pest detection, reducing crop damage, and optimizing the use of pest control resources.