Intelligent large-scale flue-cured tobacco grading based on deep densely convolutional network

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Manual flue-cured tobacco grading is inefficient, subjective, and labor-intensive.
• Existing automated methods struggle with accuracy, especially with numerous classes and limited, low-resolution datasets.
• There is a need for intelligent grading systems that can handle large, high-resolution tobacco datasets.

Purpose of the Study:

• To develop an efficient and intelligent flue-cured tobacco grading method.
• To address the limitations of existing methods in feature extraction and adaptability to multiple grades.
• To create and validate the largest, highest-resolution flue-cured tobacco dataset available.

Main Methods:

• Collected the largest and highest-resolution flue-cured tobacco dataset.
• Proposed a novel grading method based on deep densely convolutional network (DenseNet).
• Employed DenseNet's unique connectivity for enhanced feature extraction and reduced information loss.

Main Results:

• The proposed DenseNet model achieved an accuracy of 0.997 in flue-cured tobacco grading.
• Demonstrated superior performance compared to traditional and other intelligent grading methods.
• Verified the usability of the new dataset through experiments with various algorithms.

Conclusions:

• The DenseNet-based approach offers a highly accurate and efficient solution for flue-cured tobacco grading.
• The developed high-resolution dataset is valuable for training and validating advanced grading models.
• This intelligent system overcomes the limitations of manual grading and prior automated techniques.