Microbial Colony Species Recognition Using an Enhanced YOLOv4 Algorithm with CBAM and k-means++ Optimization

Abstract

The identification of microbial colonies based on their characteristics, such as size, shape, and color, is essential for recognizing different species, especially in the field of food hygiene. Traditional manual identification methods are labor-intensive and prone to error. This paper proposes an improved microbial colony recognition algorithm based on YOLOv4 to address these challenges. The method incorporates image preprocessing steps like median filtering and enhancement, followed by target labeling using the labelimg tool. The original YOLOv4 network structure is enhanced by introducing the Convolutional Block Attention Module (CBAM) to filter redundant information and using the k-means++ algorithm to optimize anchor boxes for improved object detection. The proposed model achieves a 9% increase in recognition accuracy over the unmodified YOLOv4 network in identifying Escherichia coli colonies. The results demonstrate the algorithm’s high efficiency and accuracy compared to traditional manual methods, highlighting its potential for broad applications in microbial detection.