Arecanut Disease Prediction Using MobileNet V2, ResNet and VGG-16 Framework

Arecanut is one among the important plantation in tropical region. During its life cycle., it is infected by a wide variety of diseases impacting the quality and the quantity of yield, from root to fruit. India stands in second position in yielding and consuming areca nuts in the world. The existing approach for identifying disease is by simple examination by naked eye, farmers must timely analyse each crop to detect the disease. In this paper, a model is proposed for detecting diseases of areca nut leaves, stem and its trunk using three convolutional neural network architecture: Mobilenet V2, VGG16 and ResNet which are known for their characteristics, learning pattern, image classification offering light weight network design. The dataset considered is a combination of four distinct conditions: Mahali disease, stem bleeding, ring spot and yellow leaf spot. The model take image as input is pre-processed through reshaping and resizing based on the requirement, determine learnable weight and distinguish objects in the image and learn from the results. The model is compiled using categorical cross-entropy as the loss function with adam as optimizer function and accuracy as metrics. The results of the proposed model VGG-16 with strong accuracy rate of 88%, lightweight MobileNet-V2 architecture with lower computational resources also provides competitive accuracy of 87% and ResNet provides stability for both training and validation accuracy of 92 %.