Development of an Optimized Extreme Learning Machine for Rice and Maize Disease Identification System

To effectively control crop diseases and avoid agricultural productivity losses, accurate and fast crop disease detection is crucial. Conventional disease detection techniques are frequently tedious and subjective, emphasizing the need for automated and intelligent systems. Due to their quick training times, strong generalization capabilities, and capacity to handle high-dimensional data, Extreme Learning Machines (ELM) have become a potential technique for a variety of agricultural applications. A large dataset of maize and rice plants, both healthy and ill, was collected. Each image in the collection, which featured plants with various illnesses, had a label in the dataset. The puma-optimized extreme learning machine (PO-ELM) model was trained using MATLAB and supervised learning techniques after the pictures were pre-processed and the features were retrieved using principal component analysis (PCA) in order to effectively detect the disorders. To optimize the PO-ELM model's performance, the feature selection process that helped reduce the input dimensions and eliminate unnecessary features was used. The study's findings indicate that, in comparison to traditional ELM, the PO-ELM model consistently exhibits lower false positive rates over a range of thresholds, greatly reducing the rate of inaccurate disease identifications. According to the tables, PO-ELM outperforms ELM in key performance parameters. The false positive rate for PO-ELM drops significantly, from 1.78% at a threshold of 0.3 to 1.25% at 0.75.  The PO-ELM is recommended for the detection of rice diseases such bacterial blight, blast, brown spot, and tungro, and for the detection of maize diseases including blight, common rust, gray leaf spot, and healthy plants.