Deep Convolution Self-Attention Based Cascaded Autoencoder Network for Plant Disease Detection

Abstract

Early detection and treating the disease are the major need to enhance the growth and yield of agricultural plants. In general, manual monitoring of plant diseases will not provide exact outcomes. Thus, it is crucial for developing an automated system for detecting and classifying varied types of plant diseases. In recent studies, several models were utilized for classifying the varied categories of plant diseases in different species. However, the existing studies faced various limitations such as higher over fitting issues, computational complexity, reduced learning ability, etc. Therefore, the proposed study aims to develop an efficient hybrid deep learning model for enabling plant disease classification.

The steps that involved in this study are image acquisition, pre-processing, feature extraction, and classification. Initially, the input samples are gathered from the publicly available dataset containing images of diseased plants from various species including Apple, Blueberry, Cherry, Corn, Grape, Orange, Peach, Pepper, Potato, Raspberry, Soybean, Squash, Strawberry, and Tomato. Then, pre-processing is done to eliminate the noises and enhances the quality of inputs using Upgraded Gradient-based Guided Filtering (UG_GFil) method. For segmenting the affected portions in the diseased plant, Upgraded k-means clustering is used. From the segmentation samples, the needed features are extracted by introducing a novel Feature Attentional EfficientNetB0 model. After feature extraction, the relevant features are selected using Tuna Swarm Optimization. Then the types of diseases from varied species of plants are categorized by proposing a new Deep Convolution Self-Attention based Cascaded Autoencoder Network (DCSA_CAEN) model. Also, the parameters of proposed classifier are optimally tuned by using Binary Kookaburra Optimization (BKO) approach. Thus, the proposed study effectively classifies the presented plant diseases from the given input dataset.