Deep Learning based Rice Plant Disease detection and classification using Densely Convolution Neural Network (DenseNet) with Multi-Layer Perceptron (MLP)

Abstract

Rice is a staple food crop for more than half of the world's population, playing a crucial role in global food security. However, the cultivation of rice is frequently threatened by a range of diseases, ranging which can negatively affect food supply chains and result in large yield losses. In order to ensure sustainable development of rice and secure food resources, it is crucial to understand and deal with rice plant diseases. Rice plant diseases can be broadly classified into fungal, bacterial, viral, and nematode diseases. Among these, fungal diseases are the most prevalent and destructive, with rice blast, sheath blight, and brown spot being particularly notorious. In recent years, the rapid advancements in deep learning have opened new avenues for addressing complex problems in various domains. This paper introduces a novel deep learning-based model for the detection and classification of Rice Plant Disease (RPD) using a combination of Densely Convolutional Neural Network (DenseNet) and Multi-Layer Perceptron (MLP), termed the DenseNet-169-MLP model. The input consisted of pictures of afflicted rice leaves on a white backdrop. Following the required preprocessing and cleaning of the collected data which may involve fixing missing values, harmonizing the data format, as well as getting rid of noise. Find pertinent characteristics in the dataset that can be used to distinguish between various illnesses. This involves techniques such as image processing to extract  features from images of affected rice plants. Based on the anticipated disease class, group rice diseases into several categories. This can help with identifying disease trends, putting suitable management plans into place, and giving farmers focused advice. A variety of Deep Learning techniques were used to train the dataset including AlexNet,VGG16 and DenseNet-169-MLP. DenseNet-169-MLP achieved an accuracy of 94.05 when applied on the Rice plant disease dataset.