A deep semantic network-based image segmentation of soybean rust pathogens.

Area of Science:

• Plant Pathology
• Mycology
• Computational Biology

Background:

• Asian soybean rust, caused by *Phakopsora pachyrhizi*, is a significant threat to soybean production, potentially causing up to 80% yield loss.
• Accurate image segmentation of fungal pathogens is crucial for understanding disease progression and facilitating the development of new agricultural biocides and medicines.

Purpose of the Study:

• To develop an enhanced image segmentation method for accurately identifying and analyzing *Phakopsora pachyrhizi* in high-density conditions.
• To improve the efficiency and accuracy of large-scale phenotypic screening for plant fungal pathogens.

Main Methods:

• An improved Mask R-CNN model was developed, incorporating Res2net for enhanced feature extraction and FPG for improved capability.
• The model's loss function was optimized using the CIoU loss function for boundary box regression, accelerating convergence and improving classification accuracy.

Main Results:

• The enhanced algorithm demonstrated significant improvements in mean average precision (mAP) for detection (6.4%) and segmentation (12.3%) compared to the original Mask R-CNN.
• The accuracy of the improved algorithm saw a 2.2% increase, indicating superior performance in segmenting dense fungal images.

Conclusions:

• The proposed method offers a more suitable and effective tool for the precise segmentation of fungal images, particularly for plant pathogens like Asian soybean rust.
• This advancement supports large-scale phenotypic screening, accelerating research in discovering new fungicides and medicines.