Author 1 :- SUNEETA DEVI ( Research Scholar )
Author 2 :- DR. SAKULDEEP SINGH ( Asociate Professor )
The detection of fruit diseases is essential and must be conducted promptly to improve agricultural output and mitigate crop losses. In this regard, a classification of fruit diseases using CNN, enhanced by effective transfer learning, is presented. The pre-trained weights for MobileNetV2 and VGG16 models are used, with some initial layers selected frozen to balance model performance and computational economy. This method will enable us to preserve essential characteristics acquired from large datasets while minimizing training demands on constrained hardware. Optimizing the model may provide high classification accuracy while decreasing processing time and reducing RAM use, hence making the method ideal for deployment on resource-constrained devices. To enhance variability in the dataset and mitigate overfitting, augmentations such as rotation, flipping, and zooming will be applied to the enhanced data. Experimental sessions were conducted using a publicly accessible collection of fruit disease photos from several categories, including both healthy and sick states. The findings unequivocally demonstrate that MobileNetV2 provides the optimal balance between accuracy and efficiency for real-time applications. This study demonstrated an effective methodology for detecting fruit illnesses using lightweight transfer learning models and offered valuable insights for utilizing this technology in precision agriculture within resource-limited environments. Keywords: Transfer Learning, Data Augmentation, Fruit Disease Detection, Lightweight Models
