Using Deep Neural Networks to Evaluate Leafminer Fly Attacks on Tomato Plants

Among the most common and serious tomato plant pests, leafminer flies (Liriomyza sativae) are considered one of major tomato-plant-damaging pests worldwide. Detecting infestation quantifying severity these essential for reducing their outbreaks through effective management ensuring successful production. Traditionally, detection quantification performed manually in field. This is time-consuming leads to inaccurate protection practices owing subjectivity evaluation process. Therefore, objective this study was develop a machine learning model automatic estimation leaf symptoms fly attacks. The dataset used present comprised images pest on leaves acquired under field conditions. Manual annotation classify into three groups: background, leaf, from flies. Three models four different backbones were compared multiclass semantic segmentation task using accuracy, precision, recall, intersection over union metrics. A comparison results revealed that U-Net with Inceptionv3 backbone achieved best results. For symptom severity, FPN ResNet34 DenseNet121 backbones, which exhibited lower root mean square error values. computational proved promising mainly because capacity automatically segment small objects captured challenging lighting conditions complex backgrounds.