An improved deep learning network for image detection and its application in Dendrobii caulis decoction piece

Area of Science:

• Computer Science
• Artificial Intelligence
• Pharmacognosy

Background:

• Accurate identification of Dendrobii caulis decoction pieces is crucial due to high demand for quality products.
• Current identification methods are complex, limiting accessibility for ordinary consumers.
• Need for automated, consumer-friendly methods for Dendrobium species identification.

Purpose of the Study:

• To develop an improved deep learning model for accurate identification of Dendrobii caulis decoction piece species from images.
• To enhance feature extraction capabilities for detecting dense and small targets in Dendrobium images.
• To provide a reliable tool for consumers to assess product quality.

Main Methods:

• Proposed an improved YOLOv5 deep learning network incorporating a C2S module for enhanced feature extraction.
• Integrated Reparameterized Generalized Feature Pyramid Network (RepGFPN) and Optimal Transport Assignment (OTA) for improved feature fusion.
• Established a new large-scale dataset of Dendrobium images for training and validation.
• Evaluated the model's performance against existing methods using mAP@.05 metric.

Main Results:

• The improved YOLOv5 model achieved a high detection accuracy with an average mAP@.05 of 96.5%.
• The C2S module effectively improved feature extraction for small and dense targets.
• RepGFPN and OTA modules enhanced the integration of multi-dimensional features.
• The model demonstrated superior accuracy compared to other models with similar computational complexity.

Conclusions:

• The improved YOLOv5 model offers a highly accurate and efficient solution for identifying Dendrobii caulis decoction piece species.
• The developed deep learning approach can assist consumers in making informed purchasing decisions.
• This research contributes to the quality control of Dendrobium products through advanced image recognition technology.