Maize quality detection based on MConv-SwinT high-precision model.

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Traditional corn quality detection relies on subjective human inspection, leading to high error rates.
• Automated methods are needed to improve the accuracy and efficiency of corn quality assessment.

Purpose of the Study:

• To develop and evaluate an enhanced Swin Transformer model for accurate corn quality classification.
• To integrate machine vision and deep learning for objective corn quality assessment.

Main Methods:

• Collected and preprocessed 20,152 images of high-quality, moldy, and broken corn.
• Employed a Swin Transformer base model, extracting and fusing shallow and deep image features.
• Utilized a specialized convolutional block and an attention layer for feature processing and classification.

Main Results:

• The proposed MC-Swin Transformer model achieved a recognition accuracy rate of 99.89%.
• Demonstrated superior performance over traditional convolutional neural network models in accuracy, precision, recall, and F1 score.
• The model effectively and efficiently classifies different corn qualities.

Conclusions:

• The MC-Swin Transformer offers a novel and effective technical approach for automated corn quality detection.
• This advancement has significant implications for improving smart agriculture practices.
• The study highlights the potential of deep learning in enhancing agricultural product quality assessment.