Robust method for classification of agricultural crops diseases using LGXP descriptor

• The LGXP descriptor effectively captures crop leaf disease patterns with robustness. • MRMR reduces feature redundancy while retaining key discriminative patterns. • K-fold cross validation with SVM kernels affirms stable and reliable classification. • The framework supports strong generalization across diverse crop datasets. The productivity of agriculture needs to be increased to meet the global demand driven by population growth. As a result of environmental degradation and loss biotic & abiotic agents, several kinds of diseases affect agricultural crops and drastically reduce yield. Early diagnosis of the crop diseases is crucial for applying manures and pesticides in a controlled manner without harming the environment. Researchers have developed various methods for identifying crop diseases, right from conventional techniques to the latest machine learning (ML) and deep learning (DL) techniques. However, these methods often suffer from reduced accuracy and high computational complexity. To address these challenges and to design a framework that is both efficient and suitable for real-time hardware implementation, an effective feature extraction and classification strategy has been adopted. In the proposed method, local Gabor XOR pattern (LGXP) features are extracted by applying multi-scale, multi-orientation Gabor filters followed by XOR-based neighborhood encoding, which captures discriminative texture variations associated with disease symptoms. For classification, support vector machines (SVM) with different kernel functions (such as: linear, quadratic, cubic, and Gaussian) are employed to ensure robust decision boundaries across diverse datasets. Among various kernels, linear and coarse Gaussian resulted in better disease classification performance. This method has been validated on multiple crop disease datasets, including paddy, wheat, maize, and mango, achieving the highest accuracy, sensitivity, specificity, F1-score, and MCC of 100%, 100%, 100%, 100%, and 1, respectively for cereal crops. The proposed approach demonstrates robustness in terms of computational efficiency and improved classification performance across multiple disease classes.