A multi-view multi-label fast model for Auricularia cornea phenotype identification and classification.

Area of Science:

• Agricultural Science
• Computer Vision
• Biotechnology

Background:

• Accurate identification and classification of Auricularia cornea fruit body phenotypic features are essential for quality grading and breeding.
• Phenotypic features like size, shape, color, and damage are challenging to classify rapidly due to their distribution across multiple views.

Purpose of the Study:

• To develop a novel multi-view, multi-label fast network for simultaneous identification and classification of six phenotypic features of Auricularia cornea.
• To improve the efficiency and accuracy of phenotypic feature extraction and classification for Auricularia cornea.

Main Methods:

• A multi-view feature extraction model using partial convolution and channel attention mechanisms was developed.
• An efficient multi-task classifier based on class-specific residual attention was designed.
• Task weights were dynamically adjusted using heteroscedastic uncertainty to reduce training complexity.

Main Results:

• The proposed network achieved a classification accuracy of 94.66% on a dataset of dried Auricularia cornea.
• The network demonstrated a fast inference speed of 11.9 ms.
• The system successfully identified and classified six phenotypic features simultaneously from three different views.

Conclusions:

• The novel multi-view, multi-label fast network enables efficient and accurate identification and classification of Auricularia cornea phenotypic features.
• This approach has significant potential for applications in mushroom quality grading and breeding programs.
• The integration of attention mechanisms and dynamic task weighting enhances classification performance and reduces computational complexity.