Breast Cancer Identification Using Improved DarkNet53 Model

Breast cancer is a challenging problem in the field of Bio-Medical image processing. In this research, our focus is to improve the accuracy of multiple datasets. In this proposed framework multiple pre-trained networks have been used to predict cancer. We use Darknet53, SqueezeNet, and ResNet50 on two different datasets of pathological images. The goal is that experimental shape elucidation requires substantial effort and can be time-consuming. As a result, we seek an automatic technique for detecting cancer. The increased number of softmax classifiers, leakyrelu activation layer, and additional Batch normalization layer is implemented in darknet53 mod-el. As a result, changes aided in the improvement of the Darknet model's structure and parameters. These findings show that this technique can extract multi-layer characteristics from cancer pathological images efficiently and accurately, regardless of batch size. In our research, we added an additional batch normalization layer to all three used networks, which improve the learning rate, and validation accuracy and made learning easier. Among used networks, Darknet53 achieved the highest accuracy with 95%.