The purposed ARUNet6L model is a deep learning-based hybrid model for performing image segmentation of mitochondria efficiently and accurately. The ARUNet6L model constitutes of U-Net architecture, residual connection at encoder side to improve gradient flow and resolve the problem of vanishing gradient. Attention mechanism attached to skip connection for transferring the most relevant information to the decoder and ignore the irrelevant background noise. Transpose convolution is applied for precisely reconstructing the images at the decoder side. The proposed model ARUNet6L is six layers deep for extracting fine structural details of mitochondria and Dice Loss function is used for managing imbalance in mitochondria images. The experimental results show that the ARUNet6L model outperformed the existing U-Net models and their variants. The performance was evaluated on the metrices like F1 Score: 0.9320, Mean Dice Coefficient: 0.8398, IoU score: 0.8789, Sensitivity: 0.8181 and Specificity: 0.9073. The proposed model is able to segment images of mitochondria efficiently. High F1 score indicates that there is huge similarity between the true mask and the mask predicted by the model. The ARUNet6L model is a variant of U-Net that performs effective feature extraction.