Fine-tuning and continuous training of agricultural disease classifier
Fine-tuned a MobileNetV3 model on a custom dataset of 28,000 crop leaf images across 12 disease classes for disease classification. The model was retrained regularly using new, labeled feedback data validated by farmers. Labeled data was used bi-weekly to ensure continuous improvement of the classifier for agricultural disease detection. • Curated and annotated a custom dataset of crop leaf images with class labels indicating specific diseases. • Built and deployed a feedback loop system allowing end users to validate and submit additional label corrections. • Regularly processed new data from the feedback loop for bi-weekly fine-tuning sessions. • Achieved high accuracy in real-world diagnosis use cases for farmers.