| Popis: |
The article discusses the recognition of pathologies on X-ray images using convolutional neural networks. To improve the quality of the model, we used training at different speeds of different layers. As a result of a large number of experiments and enumeration of training parameters on a random grid, the best parameters were determined: the number of epochs – 50, speed, and a decrease in the learning rate by 10 times every 20 epochs. For training a fully connected layer, a speed of 0.0012 was used, and for training convolutional layers, a speed of 0.00082. As a result of the work, a model was created that predicts the area of pathology on X-rays of the chest. As a result of the work, it was determined that the best results can be achieved using the network of the ResNet architecture, at best, an accuracy of 87 % was obtained. |
