An intelligent Deep Combined Vector Optimized Learner (DCOL) for plant leaf disease identification and class prediction – Scientific Reports

Abstract

Plant leaf diseases significantly affect agricultural productivity, leading to substantial crop losses and economic impact. Early and accurate detection is essential for ensuring food security and enabling timely intervention. Existing approaches based on conventional image processing and standard deep learning models often suffer from limitations such as poor generalization, sensitivity to background variations, inadequate feature representation, and high misclassification rates in visually similar disease patterns. Moreover, numerous approaches depend on single-stage learning frameworks that inadequately capture both spatial and textural attributes of affected areas. This study introduces a hybrid framework called Deep Combined Vector Optimised Learner (DCOL) to tackle the issues of plant leaf disease diagnosis and classification. The proposed method integrates preprocessing using Nonlocal Means filtering for noise reduction, followed by segmentation using DeepLabv3 + with atrous spatial pyramid pooling to accurately isolate disease-affected regions. Texture-based features are extracted using the Gray-Level Co-occurrence Matrix (GLCM), while deep discriminative features are learned through a Deep Combined Spatial Attention Network (DCom-SAN) incorporating channel, spatial, and position attention mechanisms. These complementary features are fused and optimized using the Average Vector Weighted Optimization (AVWO) algorithm to enhance convergence and classification performance. The model is assessed using the publicly accessible Plant Village and Cotton Leaf datasets, attaining an accuracy of 99.26%, alongside enhanced precision, recall, and F1-score relative to conventional machine learning and deep learning methodologies, including SVM, VGG-16, ResNet-50, and EfficientNet. The findings indicate that the proposed DCOL framework offers enhanced feature representation, strong generalisation, and dependable disease classification, rendering it appropriate for practical use in agriculture.

Subjects