Precision Agricultural Early Stage Disease Detection In Plant Leaves Using Deep Learning

Abstract

Detecting plant disease in an earlier stage is one of the tasks where these can lead to heavy loss in cultivation. In such a way the research challenges are more likely to relate to the detection and diagnosis in plant diseases based on spots affected by many viruses that can be handled using deep learning techniques. Agricultural field relies on extracting many features from the dataset to understand the automation of identifying the diseases. Mostly machine learning is better in providing solutions for early diagnosis, even though there are many limitations such as computational cost and time. To overcome these limitations Deep learning models are introduced in identifying the diseases in plants. There is trained and tested data which are independent of classes and varieties such that image databases refer to the labeling for finding the severity. This work focuses on providing solutions by classifying, segmenting input from the images. By applying Convolutional neural network the comparison analysis with the enhanced model has been introduced. Using trained and tested images from 38 classes of various plants that were taken from Kaggle image datasets. This research implies three different models using deep learning models for finding and detecting the infection in plants. In the first research model, the plant disease is detected accurately using a deep learning algorithm such as Enhanced Mask R-Convolution Neural Network (EnMask-RCNN). Based on the image classification, segmentation is processed within the region and its boundaries using the threshold value is computed to increase the performance and accuracy of efficiency in a better manner. The proposed result is compared with the existing technique Recurrent Neural Network (RNN) which uses a sample size of 105 that has a less error rate of 2.657 which is less than RNN. The significance value proves better than 0.258 and the loss value defines the reliable results. The second model applied here is to detect the leaf infection in an earlier stage and segment