| dc.description.abstract | Aiming at the problems of low detection accuracy and slow detection speed in traditional human polyp detection, this paper proposes a colorectal polyp detection method based on improved yolov4 algorithm. Firstly, the colorectal polyp data set is constructed. In order to automatically detect the polyp position in CT and increase the accuracy, deep learning is needed to make the program remember the characteristics of polyps. The deep learning environment is built with anaconda, mainly using CSP packet as the backbone network, CT image as the data set, and using Python to train. Secondly, the training mainly uses mosaic data enhancement, learning rate cosine annealing attenuation, and the activation function uses mish activation function. Finally, if there is, output "0" to represent the presence of polyps. When the polyp is found in CT, it will be marked with a red box, and the center coordinates of the red box, as well as the height and frame degree of the red box will be output; On the basis of the above requirements. | ru |
