Improved detection and counting performance of microplastics in common carp whole blood by an attention-guided deep learning method

Microplastics are ubiquitous pollutants in the living environment and food chains. Microplastic particles were recently even discovered in human blood for the first time. For high-throughput biodistribution analysis in toxicity studies, we propose a new deep-learning-based method for the automatic identification of polystyrene particles in whole blood. Attention-YOLO helps the network improve detection accuracy by adding the channel attention mechanism to the feature extraction network. We use microscopy to collect a dataset consisting of bright-field images of particles in various colors to train and test a neural network model. Then we use the model to identify and count polystyrene particles in untrained carp blood. The experimental results show that the Attention-YOLO network, when compared to the standard YOLO network, can achieve a better detection performance of microplastics counting in carp blood without adding too many extra parameters, with improvements in the recognition accuracy of polystyrene microspheres (ps-Bs, ps-Rs, and psGs) of 0.2%, 1.1%, and 6%, respectively, and the mean Average Precision (mAP) of 2.4%.