Initial detection of breast cancer is necessary to reduce mortality and treatment expenses. This research presents a new structure to improve breast cancer detection in mammographic images. It performs through residual network (Resnet) with the respective adaptive histogram equal (AHE), expected maximum Gaussian mixture model (EMGMM) division, and learning (TL). The process starts with pre-processing through AHE to increase the image contrast. This improves the prominence of subtle characteristics in mammograms. Subsequently, EMGMM identifies and outlines areas of interest (ROI) within images. After partition, areas are classified using the Resnet. The Resnet is pre-educated at a heavy dataset and tuned for breast cancer detection tasks. TL with Resnet uses its ability to recover important features, which increases classification accuracy. The proposed research is evaluated with accurate measures such as standard matrix and accurate measures such as the association (IU) to evaluate the partition accuracy. Comparing this combined AHE, EMGMM and Resnet with existing approaches shows that it achieves better performance in initial breast cancer detection.