A DEEP LEARNING FIELD PLANT IMAGE DATABASE FOR DISEASE DETECTION AND CLASSIFICATION

Abstract

To meet the 70% expansion in worldwide food yield by 2050 expected by plant sickness losses, analysts have assembled prevalent deep learning models. These models, prepared on PlantVillage, battle in real-world field conditions because of convoluted settings and many leaves per shot. This exploration gives FieldPlant, another assortment of 5,170 plant infection photographs clarified by plant pathologists from estates. Present day order calculations like MobileNet, VGG16,

InceptionResNetV2, InceptionV3, Xception, and DenseNet are tried for tropical corn, cassava, and tomato diseases. Moreover, YoloV5, YoloV8, SSD, and FasterRCNN plant ID calculations are assessed. DenseNet and Xception succeed at order, while YoloV5 succeeds at plant discovery with 97% accuracy and 0.977 mean Average Precision. This study demonstrates the way that improved strategies can change crop illness finding and decrease overall result losses.[17