Citrus leaf disease detection by image processing

Abstract

Citrus leaf diseases bring a danger to the earnings of citrus estates. When it comes to recovering from illness, early detection and accurate diagnosis are very necessary. In the last several decades, there have been advancements made in the diagnosis and classification of citrus leaf diseases via the use of deep learning techniques in image processing. When it comes to automating the detection of citrus leaf diseases, we recommend making use of pre-trained convolutional neural networks (CNNs) like ResNet-50, VGG16, MobileNet-V2, InceptionV3, InceptionResNet-V2, DenseNet- 201, and DenseNet-121.To accomplish this goal, a comprehensive data collection consisting of images of citrus leaves that have been identified will be gathered and pre-processed. Citrus canker, greening, and black spot leaves will be included in the databases, along with healthy and diseased citrus leaves.For the purpose of extracting useful characteristics from leaf images, we shall make use of deep learning models. For the purpose of picture classification, the models that were discussed before are useful and often used.In this research, we propose a CNN model that is both effective and efficient. The model was originally trained on 596 pictures, and then it was augmented with 2800 images that were divided into three categories: training, validation, and testing.70% of the data goes towards training, 15% goes towards validation, and 15% goes towards testing.A few pieces of public data will be served this model.Following that, we will evaluate the findings in relation to the prebuilt models.Last but not least, we get a training accuracy of 95.95% and a validation accuracy of 97.84%. Furthermore, our suggested model has a lower number of training parameters compared to all other pretrained models, which enables our model to categorise illnesses more quickly.This provides us with a decent level of accuracy.