Chickpea wilt is a widespread agricultural disease that affects production worldwide every year. Rapid and accurate detection of the disease is desirable, but is difficult using traditional methods. Therefore, it is necessary to detect the disease using automatic, rapid, reliable, and simple methods before it completely damages the plant. Herein, we investigate the applicability of machine learning-based texture analysis methods to determine the severity level of Fusarium wilt in chickpea. Various procedures, such as image annotation, augmentation, resizing, and color conversion using different color spaces (RGB, HSV, and Lab*), were performed to develop the model. To perform texture feature extraction, the Gray-Level Run-Length Matrix (GLRLM) and the Gray-Level Occurrence Matrix (GLCM) feature extraction methods were used. To avoid local minima, Bayesian optimization was applied, while to train and test the effectiveness of the proposed model, 15000 images (70–20-10 ratio for training, validation and testing) were used. Finally, multi-class classification models were developed using image classification methods such as K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Neural Networks. The proposed GLRLM-HSV based KNN model performed well in determining the severity level of fusarium wilt of chickpea among five different severity levels, with an accuracy of 94.5%.

鹰嘴豆枯萎病是一种广泛传播的农业病害，每年都会影响全世界的生产。快速准确地检测该疾病是可取的，但使用传统方法很难。因此，有必要在病害完全损害植物之前，采用自动、快速、可靠、简单的方法对其进行检测。在此，我们研究了基于机器学习的质地分析方法的适用性，以确定鹰嘴豆枯萎病的严重程度。执行了各种程序来开发模型，例如使用不同颜色空间（RGB、HSV 和 Lab*）的图像注释、增强、调整大小和颜色转换。为了进行纹理特征提取，使用了灰度游程矩阵（GLRLM）和灰度出现矩阵（GLCM）特征提取方法。为了避免局部极小值，应用了贝叶斯优化，同时为了训练和测试所提出模型的有效性，使用了 15000 张图像（70-20-10 的比例用于训练、验证和测试）。最后，使用 K 最近邻 (KNN)、支持向量机 (SVM) 和神经网络等图像分类方法开发了多类分类模型。所提出的基于 GLRLM-HSV 的 KNN 模型在确定鹰嘴豆枯萎病五个不同严重程度级别的严重程度方面表现良好，准确度为 94.5%。