Enhancing leaf disease detection accuracy through synergistic integration of deep transfer learning and multimodal techniques

Abstract

The agricultural sector, a cornerstone of economies worldwide, faces significant challenges due to plant diseases, which severely affect crop yield and quality. Early and accurate detection of these diseases is crucial for effective mitigation strategies. The current methods used often lack accuracy and adaptability, especially in diverse environmental conditions. This study introduces a novel, synergistic approach that integrates deep transfer learning with multimodal techniques, specifically canny edges, colour spectrum intensity analysis, and custom data augmentation strategies. Unlike existing methods that rely solely on pre-trained models, the approach utilised in this study offers an innovative fusion of distinct feature extraction techniques. The canny edges highlighted the structural intricacies of leaf diseases, while colour spectrum intensity analysis enhanced the detection of disease-specific colour markers. The customized data augmentation techniques employed (in the study) was shown to enhance the learning process of the models, resulting in their adaptability to diverse agricultural environments. This integration applied to DenseNet201 and EfficientNetB3, achieved detection accuracies of 99.03 % and 98.23 %, respectively, surpassing traditional models and setting new benchmarks in plant disease detection. These results demonstrate the effectiveness of the proposed multi-faceted approach and its potential to significantly enhance crop disease management systems.