Determining the cause of cardiovascular illnesses and identifying cardiac-related issues depend heavily on accurate biomedical image segmentation. It primarily aids in the acceleration of illness identification because manual cardiac segmentation by physicians or audiologists can be laborious and unreliable, depending on the experience and competence of the operator. Significant advancements have been made in this field due to the quick growth of deep learning. Nevertheless, traditional automated learning algorithms are unable to divide the whole heart. In turn, the U-Net model for biomedical cardiac medical image segmentation has been applied in this research. The suggested U- Net model uses an encoder that has been enhanced using transfer learning techniques, which enables it to learn from a small quantity of input efficiently. Our primary goal is to classify the myocardium accurately, both left and right valve views in our segmented envision, while segmenting the whole heart image. This component enhances the model’s ability with additional classification and localization, providing improved precision and accuracy in both aspects. Our datasets were acquired from ACDC (Automated Cardiac Diagnosis Challenge). The model performed a dice loss of 0.0185–0.0265 all over the epochs, which is phenomenally good in the segmentation of images.