Advancing mango leaf variant identification with a robust multi-layer perceptron model.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Mangoes are vital in Bangladesh, but identifying varieties from leaves is difficult.
• Existing research primarily uses fruit images, neglecting leaf-based classification.

Purpose of the Study:

• To develop an automated system for classifying mango types using leaf images.
• To introduce a novel deep learning model, WaveVisionNet, for this purpose.

Main Methods:

• Curated and augmented the MangoFolioBD dataset with 16,646 high-resolution mango leaf images.
• Developed and validated the WaveVisionNet model, a multi-layer perceptron, on leaf image datasets.
• Evaluated WaveVisionNet against state-of-the-art models like Vision Transformer and transfer learning approaches.

Main Results:

• WaveVisionNet achieved high accuracy rates of 96.11% on a public dataset and 95.21% on the MangoFolioBD dataset.
• The model outperformed existing state-of-the-art models in mango leaf identification.
• WaveVisionNet effectively combines lightweight CNNs with noise-resistant techniques for robust analysis.

Conclusions:

• Automated mango leaf identification using WaveVisionNet offers significant benefits for farmers and agricultural stakeholders.
• The model enables precise plant health diagnosis, enhancing agricultural practices and crop quality.
• This technology supports improved crop yields and quality through early variety identification.