Computer Vision Based Automatic Leaf Disease Prediction using the Trained Convolution Base Focusing on Transfer Learning

Abstract

Plant are considered significant because they are the origin of humanity’s supply of energy. The leaf may be infected by plant diseases at any time between sowing and harvesting, leading to a huge loss in crop production and economic market value. In monitoring large fields of crops, the identification of plant diseases has now received growing attention. When moving from one disease control strategy to another, farmers face great difficulties. Experts naked eye observation is the standard approach to the detection and identification of plant diseases adopted in nature which is still in practice by many farmers. This article presents a Transfer learning techniques which is used for automatic detection and classification of plant leaf diseases. Using a public dataset of 1821 training images out of which 516 images are healthy and 1305 images are infected images. A deep convolution neural network with trained convolution base with ImageNet using ResNet, DenseNet and EfficientNet model. The selected model yield accuracy of 97% in ResNet and 98% in DenseNet. The qualified model the EfficientNet on a held-out test range achieves an accuracy of 99.5%, showing the viability of this method. Overall, a simple path toward crop disease detection using transfer learning and compared the metrics like Accuracy, Precision, Recall and F1 score for the model ResNet, DenseNet and EfficientNet.