An Effective Machine Learning Technique for Rice Leaf Disease Detection

Plant diseases, especially those that attack rice leaves, cause serious problems for farmers since they make it more challenging to provide enough food to satisfy the world's growing population. Farmers' livelihoods that rely on agriculture have been threatened by the economic losses brought on by rice illnesses. As a tragic outcome of the hardships they encounter, several farmers have taken their own lives. It is crucial to detect and identify disease-infected plants in order to execute efficient disease management strategies. The Inception V3 model is utilized for disease classification, and the Convolutional Neural Network (CNN) autoencoder is employed to reduce noise in images of rice leaves. The dataset is then expanded through data augmentation approaches. The primary objective of the initial stage is to gather photographs of rice leaves from the UCI Machine Learning Repository. Once a collection of photographs has been gathered, the images are subjected to processing using a Convolutional Neural Network (CNN) autoencoder model. This process aims to enhance the quality of the images and eliminate any unwantednoise present in them. Subsequently, data augmentation techniques are employed to increase the volume of the dataset and standardize the dimensions of the images, both of which play a role in training a robust and adaptable model. Diseases are categorized using the popular inception v3 model, which can process vast and diverse picture datasets. This method allows for the precise categorization and detection of a wide variety of rice leaf diseases with the accuracy of 98.8%.

I. Introduction

There has been a rise in the number of plant diseases caused by pathogens like viruses, bacteria, and fungi, which can cause extensive damage to many parts of a plant (above and below ground) and make early detection and effective management of these illnesses challenging. Therefore, the ability torecognize plant illnesses from photos is an exciting area of study in the agricultural sector. Many approaches exist for efficiently utilizing image processing and machine learning in the context of plant disease diagnostics. Pre-processing techniques like filtering and the CLAHE algorithm are often used in plant image analysis, asare segmentation methods like thresholding, region expansion, and clustering, and feature extraction techniques like LBP, HOG, and the Gray-Level Co-Occurrence Matrix. Rice, scientifically known as Oryza sativa, is India's most important food crop. Leaf smut, bacterial leaf blight, and brown spot are some of the most common diseasesthat cause harm to rice leaves. The outward manifestations of these diseases vary widely. However, because many diseases share similar outward appearances, it might be difficult for farmers to physically identify a specific ailment. By using image processing techniques, we may dramatically reduce labor expenses and enhance the accuracy of illness diagnosis [1]. The utilization of machine learning and image processing techniques for the automated identification of diseases are significant in mitigating the laborious and time-intensive procedure associated with manual detection. Rice is a staple food in China, although India grows more rice than that country. Odisha ranks fourth among the Indianstates in terms of rice output. Sambalpur and the neighboring district of Bargarh are regarded as the “rice bowl of Odisha” due to their prominence in the state's rice producing industry. From July to October, during the monsoon season, numerous varieties of rice are planted here; from October to March, when water is supplied by the Hirakud dam, a separate set of rice cultivars is grown. Each year, pests and diseases wreak havoc on the rice fields in this region. However, many inexperienced young farmers struggle to identify the precise illnesses ailing their crops, leading them to overuse or misuse pesticides. This circumstance emphasizes the significance of advancing knowledge about the causes of rice diseases in Odisha's western area. Diseases such as bacterial blight, blast, brown spot, and tungro have had a significant impact on rice production. Diseases are currently diagnosed through either laborious laboratory procedures involving chemical reagents or through expert visual judgment. Both approaches, however, are time-consuming and need thorough familiarity with the topic at hand. In this age of the internet and mobile technologies, programs like “Rice Doctor” and “Rice Xpert” have been developed toassist farmers. While these farmer-friendly apps were developed with the best of intentions, they still have some work to do when it comes to recognizing rice diseases [4]. In this research, we present a very precise machine-learning strategy for forecasting and categorizing three common rice diseases. In Section II, we present our background study, and in Section III, we present our approach. This study offers a potentially game-changing new method for managing agricultural diseases in Bangladesh by utilizing machine learning and image analysis to predict and classify diseases in rice plants.