Abstract:
The comparative analysis of CNN and transfer learning approaches delves into the
application of agriculture, with the specific focus on mango leaf disease detection.
The core objective of the study is to develop an effective method for early, accurate
and sustainable leaf disease detection system. With the help of convolutional neural
networks and transfer learning techniques, this study aims to analyze raw mango leaf
images to identify the certain disease. Raw images were collected from local mango
orchards that undergo processing and data augmentation to refine model training. The
study employs an array of deep learning models, including bespoke CNN models and
pre-trained models such as EfficientNetB4 and InceptionV3. These models are
conscientiously trained and tested against the dataset, with their performances
evaluated on metrics like accuracy, precision, recall, and F1-score. It was observed
that, the CNN-02 model, designed with batch normalization and dropout layers,
exhibited superior performance in terms of accuracy and generalizability compared to
other models, achieving 95.65% of accuracy. Additionally, the EfficientNetB4 model
demonstrated an impressive learning capacity, with a precision rate of 98.3%. These
results substantiate the effectiveness of both CNN and transfer learning approaches in
the realm of leaf disease detection, with certain models showing exceptional accuracy
and efficiency and promising potential of CNN and transfer learning techniques in
early mango leaf disease detection, making a significant contribution to agricultural
technology.
