Published Paper Details:

Accurate and early detection of crop nutrient deficiency symptoms from leaf images is crucial for global food security and sustainable agriculture. Traditional methods like lab testing or manual inspection to detect crop disorders are time-consuming and costly. Furthermore, these methods fail to detect crop disorders due to the variability in field conditions. To address this challenge, we propose a novel Physiology-Guided Attention Network (PGA-EffNetB0) for automatic detection of crop leaf disorders. The proposed method initially applies botanically inspired preprocessing to each input image to extract the physiological and morphological traits of leaves, such as chlorophyll distribution, venation (leaf vein patterns), and pigmentation uniformity, which helps in transforming the input image into a more biologically informative representation. Then, a pre-trained EfficientNetB0, a Convolutional Neural Network (CNN) known for strong image classification performance with fewer parameters, utilising these informative features, was utilised to build the model. Furthermore, the architecture is enhanced with a spatial attention module to make the model more biologically aware by focusing on the most informative regions of an image rather than treating all features equally. The proposed model was trained and evaluated on the publicly available PlantVillage dataset, which comprises approximately 54,300 leaf images across 38 disease or disorder and healthy classes covering major crops such as tomato, potato, apple, maize, and grape. The proposed model attained a classification accuracy of 98.64 %, precision of 98.51 %, recall of 98.43 %, F1-score of 0.985, and a macro-averaged ROC-AUC of 0.992 on the validation set. Compared with conventional image-only CNN baselines such as ResNet50 (95.4 %) and VGG16 (94.8 %), the proposed approach improved accuracy by approximately 3-4 % and reduced misclassification. These findings confirm that integrating domain-specific botanical cues into deep networks enhances the performance and robustness that enables it to deploy in mobile or edge-based devices for sustainable crop management.