Block-based compressive sensing in deep learning using AlexNet for vegetable classification.

Area of Science:

• Computer Vision
• Machine Learning
• Agricultural Technology

Background:

• Vegetable classification is crucial for agricultural automation.
• Deep learning, specifically Convolutional Neural Networks (CNNs), offers advanced image recognition capabilities.
• Existing methods may face challenges with computational efficiency and storage.

Purpose of the Study:

• To propose an optimized vegetable classification technique using the AlexNet CNN model.
• To integrate compressive sensing (CS) with AlexNet to improve efficiency.
• To evaluate the performance of the proposed method in terms of accuracy and compression.

Main Methods:

• Utilized the AlexNet deep learning model for vegetable image classification.
• Applied compressive sensing (CS) with discrete cosine transform (DCT) for sparsing, Gaussian distribution for sampling, and orthogonal matching pursuit (OMP) for reconstruction.
• Implemented a block-based CS approach integrated with AlexNet.

Main Results:

• The standalone AlexNet model achieved a maximum test accuracy of 98%.
• The combined block-based CS and AlexNet method reached a maximum accuracy of 96.66% with a 2x compression ratio.
• The integrated method demonstrated high performance in classifying four types of vegetable images.

Conclusions:

• AlexNet CNN architecture provides robust vegetable image classification.
• Integrating block-based compressive sensing with AlexNet effectively reduces computational time and storage space.
• The proposed hybrid method offers a superior approach for vegetable classification in agricultural applications compared to previous techniques.