Classification of Maize (Zea May L) Leaf Diseases Variants Based on Sobel Edge Detection and Machine Learning Technique

Abstract

Zeae-maydis, also known as maize gray leaf spot, and porcinia sorghi, known as maize common rust, are the two most prevalent and dangerous diseases that harm maize crops in Nigeria. Plant diseases are difficult for Nigerian farmers to recognize correctly, and it is impossible to assess their severity with the unaided eye. However, hiring a pathologist is more costly and time-consuming for large farms. Moreover, many support vector machine (SVM) classification models for maize leaf disease classification have been developed by different researchers. However, these existing models are impacted by imbalanced datasets, irrelevant feature selection, and difficulty in fine-tuning the hyperparameters of the SVM. Consequently, to resolve these problems, two optimized multiclass support vector machine classification models (BPSO-SVM and RSA-SVM) were trained to categorize maize leaves disease into Zeae-maydis and porcinia sorghi using 1,648 photos of maize leaves across all maize datasets, which included 574 photos of gray leaf spot disease, 574 photos of common rust disease, and 500 photos of healthy leaves obtained from the Kaggle village datasets. The images were scaled down, converted to grayscale, and enhanced using morphological filtering, bi-histogram equalisation techniques, and adaptive median filtering before the affected area was segmented through the Sobel edge detection method. The Gray Level Spatial Dependence and colour moment were then used to extract texture, shape, and colour features, which were then fused using the linear combination method. The 10-fold approach was used to train and test each classification model. The comparative experiments demonstrate that the BPSO-SVM model outperforms the RSA-SVM model at a threshold value of 0.80. The RSA-SVM model has a performance accuracy of 95.62% and 95.25% on the datasets for gray leaf spot and common rust disease, respectively, while the BPSO-SVM has a performance accuracy of 96.37% and 96.93% on the same datasets. The two models can be used to classify Zeae-maydis and porcinia sorghi in maize, according to a comparison with the current models. However, this study only identified two of the numerous diseases that affect maize, and it offered no suggestions for how to prevent any of these illnesses.