CBIA-Net for Rapid Detection of Typical Wound Bacteria Using Hyperspectral Imaging

Bacterial infection in wounds is one of the critical factors in delayed healing and can lead to life-threatening sepsis in severe cases. The rapid and nondestructive detection of wound bacteria and timely intervention and treatment are of great significance. Hence, a novel CNN-BiGTrans interactive aggregation network, CBIA_Net, is proposed for bacteria detection using hyperspectral imaging (HSI). Specifically, two feature extraction branches BiGTrans_Branch and CNN1D_Branch are designed, the former captures the sequential information and long-distance dependencies of long sequences while acquiring global information, and the latter adequately extracts the key local features of sequences and expands the receptive field. In addition, a cross-branch weighted fusion module (CWFM) is designed to interactively fuse global and local features of spectral sequences to obtain diverse and robust features. Finally, a hyperspectral wound-typical bacteria dataset is constructed to evaluate the performance of the proposed method. The experimental results show that the proposed method has better classification performance compared with other models, reaching 97.04%, 97.03%, 97.03%, and 97.03% in the four metrics of Precision, Recall, Accuracy, and F1-score, respectively. We believe that this method can be used for the rapid detection of other bacteria and has a great potential application in the task of identifying typical bacteria on real wounds.