Overlapped tobacco shred image segmentation and area computation using an improved Mask RCNN network and COT...

Area of Science:

• Computer Vision
• Machine Learning
• Agricultural Technology

Background:

• Accurate classification and component area calculation of tobacco shreds are crucial for determining blending ratios and overall quality.
• Complex physical characteristics and overlapping/stacking of tobacco shreds, especially expanded tobacco silk and tobacco silk, pose significant challenges for traditional machine vision systems.
• Existing methods struggle with distinguishing similar varieties and accurately calculating areas in overlapped tobacco shred images.

Purpose of the Study:

• To develop a novel segmentation model for identifying various types of overlapping tobacco shreds.
• To accurately calculate the areas of these overlapping tobacco shreds.
• To address the limitations of current machine vision techniques in tobacco shred analysis.

Main Methods:

• An improved Mask region-based convolutional neural network (RCNN) was developed, replacing its backbone with Densenet121 and U-FPN.
• Region proposal network (RPN) parameters (anchor size and aspect ratios) were optimized for better feature extraction.
• A novel algorithm for calculating the overlapped tobacco shred region (COT) area was proposed and applied to segmented images.

Main Results:

• The developed model achieved a final segmentation accuracy of 89.1% and a recall rate of 73.2%.
• The average area detection rate for 24 overlapped tobacco shred samples improved significantly from 81.2% to 90%.
• The study demonstrated high accuracy in both segmentation and overlapped area calculation for tobacco shreds.

Conclusions:

• The proposed improved Mask RCNN model offers an effective solution for the type identification and component area calculation of overlapped tobacco shreds.
• This research provides a new methodological approach for similar image segmentation tasks involving complex, overlapped objects.
• The findings contribute to advancing automated quality inspection systems in the tobacco industry.