Application and System Study of Rotational Reconstructed Convolutional Neural Network: A Study on the Application System of Convolutional Neural Network of Rotating Capsule Type

At present, the trained Convolutional Neural Network model cannot recognize objects with large differences in displacement in the process of object recognition due to the variability of spatial location relationships of objects. To address this problem, this study proposes a reconfigurable rotating capsule-type Convolutional Neural Network. Firstly, in the process of Convolutional Neural Network training and recognition, all images are pre-processed in advance with background removal and planting to enhance the image evidence, and then one image refinement mask is specified for each type of training gallery, and finally the images to be recognized are rotated with each mask separately in a loop to calculate the overlap rate, and the The rotation angle corresponding to the maximum overlap rate is obtained, and the rotation reconstruction process is performed on the images to be recognized. Through the cell recognition test, the accuracy of the improved algorithm is obtained to be significantly improved, and the efectiveness of the algorithm is confirmed.