RGB Color Space-Enhanced Training Data Generation for Cucumber Classification.

Area of Science:

• Agricultural technology
• Computer vision
• Machine learning

Background:

• Cucumber classification is expertise-dependent, limiting untrained labor participation.
• Peak harvest seasons demand efficient, scalable cucumber sorting solutions.
• Current manual classification methods are time-consuming and prone to inconsistency.

Purpose of the Study:

• To develop an automated cucumber classification system accessible to individuals without prior expertise.
• To enhance the accuracy and efficiency of cucumber sorting in agricultural settings.
• To leverage deep learning for objective and consistent quality assessment of harvested produce.

Main Methods:

• A convolutional neural network (CNN) with 11 layers (2 convolution, 2 pooling, 3 dense, 4 dropout) was designed for image-based classification.
• A novel data augmentation technique embeds cucumber physical attributes (length, bend, thickness) into the RGB color space of training image backgrounds.
• Performance was evaluated using standard multi-class classification metrics (accuracy, recall, precision, F-measure) against real-world cooperative criteria.

Main Results:

• The proposed method, embedding attribute information into RGB backgrounds, achieved 79.1% classification accuracy.
• The baseline method, without RGB color space embedding, achieved 70.1% accuracy.
• The enhanced method demonstrated a 1.1 times performance improvement over the conventional approach, validating the effectiveness of the background encoding technique.

Conclusions:

• Embedding physical cucumber attributes into image backgrounds via RGB color space significantly improves CNN-based classification accuracy.
• The developed system offers a practical solution for automating cucumber classification, reducing reliance on expert knowledge.
• This approach has the potential to streamline agricultural post-harvest processes and improve overall operational efficiency.