Apple quality identification and classification by image processing based on convolutional neural networks.

Area of Science:

• Computer Vision
• Machine Learning
• Agricultural Technology

Background:

• Accurate apple quality identification is crucial for the fruit industry.
• Existing methods struggle with complex image disturbances, impacting classification accuracy.
• Automated grading systems require robust models capable of feature extraction.

Purpose of the Study:

• To develop a novel Convolutional Neural Network (CNN) model for accurate and fast apple quality identification and classification.
• To address challenges posed by complex background disturbances in real-world apple images.
• To improve upon existing methods in learning high-order, related features across network layers.

Main Methods:

• Proposed a novel Convolutional Neural Network (CNN) model designed for image-based apple quality assessment.
• The model captures specific, complex, and useful image characteristics for detection and classification.
• Employed a training and validation strategy, followed by testing on an independent dataset.

Main Results:

• Achieved a best training accuracy of 99% and validation accuracy of 98.98%.
• Demonstrated an overall test accuracy of 95.33% on an independent dataset of 300 apples.
• Outperformed Google Inception v3 and traditional methods (HOG, GLCM, SVM) in accuracy and training time.

Conclusions:

• The proposed CNN model offers superior performance for apple quality identification and classification compared to existing approaches.
• The model's ability to learn intricate features makes it highly effective even with challenging image backgrounds.
• This research highlights the significant potential of the developed model for practical applications in automated apple quality detection and grading.