Deep Neural Network-Based Approach for Small-Size Colony Counting in the Textile Industry

This study proposes a multi-scale feature map fusion method based on deep neural networks (DNNs) to achieve accurate Small-size colony counting under complex backgrounds and microscopic conditions, with a specific focus on textile-related microbial detection. By constructing a DNN model, the method performs feature extraction and analysis on small-sized colony images derived from textile samples, employing a convolutional feature fusion algorithm to enhance the recognition capability for small-target colonies— a critical need in textile antibacterial performance assessment. Experimental results, validated using a large number of textile microbial samples, indicate that, compared with the traditional manual visual counting method, the proposed approach shows no statistical significance in counting differences, while achieving higher precision and meeting reproducibility requirements. The method offers high efficiency, traceability, and repeatability, enabling rapid colony counting with simultaneous saving of results and images for subsequent verification, thus greatly optimising the textile quality control process. This approach can substantially improve counting accuracy, reproducibility, and automation, fully meeting standard detection requirements for antibacterial textiles and providing strong technical support for textile microbial analysis. Moreover, while initially developed for textile applications, it also holds potential for other industrial microbial detection scenarios.