Image processing–based stage-wise visualization and severity assessment of coconut leaf diseases

The timely identification of coconut leaf diseases is essential to reduce the loss of yields and be able to intervene in precision agriculture. The present paper suggests an Earlier Disease Diagnosis in Coconut Leaves (DD-CL) model to add together transfer learning-based convolutional neural networks and symptom progression analysis to attain precise disease identification and severity prediction at the initial stages. The model uses a pre-trained VGG16 architecture with a SGD and Adam optimizer and is trained on a Kaggle coconut leaf dataset with five classes, inclusive of: healthy, flaccidity, yellowing, drying of leaflets and CCI-leaflets. To support robustness and minimize overfitting, the images are 300 × 300 pixel, normalized and augmented. In addition to classification of the disease, DD-CL does severity estimation by examining changes in the symptoms of flaccidity to uneven yellowing and tip browning. Experimental analysis shows that the highest classification accuracy is 97.3, and precision, recall, and F1-score are 97.3, 96.5 and 96.9, respectively, which is higher than ResNetV2 and MobileNet by 3–5%. The analysis of confusion matrices proves that there is minimum misclassification between the visually similar stages of the disease, especially during early disease transitions. The suggested DD-CL framework offers a scalable, interpretable, and reliable solution to automated surveillance of coconut disease and the severity at an early stage. N performs better than the single models, thereby raising accuracy and an f-measure.