Automatic Hospital Inspection System Based on Convolutional Neural Network

To solve the problem of timely replacement of hospital infusion bottle, first, ResNet and DenseNet were utilized as basic network to detect and locate their positions. Second, transfer learning algorithm was employed for the training. Third, part of the convolutional layers were trained while the front some layers were froze, this could reduce greatly the workload of training without significantly reducing the detection accuracy; Last, Canny algorithm was applied to detect the edge information of the sling, the lever position of the liquid medicine was determined by contrasting adjacent frames of dynamic change of edge data. Reminder mechanism was triggered as the liquid lever reached the given threshold.