基于卷积神经网络的采血管铝箔帽状态检测方法OACSTPCD

Objective To propose a aluminum foil cap state detection method for blood collection tubes based on convolutional neural networks,to realize the recognition of the state of the aluminum foil cap,aiming at the challenges of high identification accuracy and speed requirements,diverse types of blood collection tubes,complicated state of aluminum foil cap state detection,and the interference from liquid on tube walls in the automated biochemical immunoassay pipelines within medical laboratories.Methods Firstly,a lightweight model design approach was adopted,which reduced the depth of the model to decrease the number of parameters and computational requirements.Additionally,channel attention mechanism was introduced to enhance the feature extraction capability of the samples.Moreover,Focal Loss was used to address the problem of mining difficult samples,further optimizing the model's performance.Finally,a teacher-student network was trained to perform knowledge distillation,resulting in the final lightweight and compact model.Results The detection method due to the lightweight design of the student network model was suitable for edge computing devices with limited resources.The parameter number of the model was only 0.354 M,the computation amount was 0.165 GFlops,the recognition speed of the Jetson Nano device was 3.42 ms,and the recognition accuracy reached 100%in the case of complex collection of blood vessels.Conclusion This study fully validates the lightweight,efficient,and practical nature of the model,indicating that the detection method based on a lightweight convolutional neural networks model can accurately identify the status of blood collection tube aluminum foil cap.It has become a solution for detecting the status of blood collection tube aluminum foil caps in the automated biochemical immunoassay pipelines within medical laboratories.