The underwater image restoration task targets retaining the true colors of the underwater scenarios captured in the images by rectifying the distorted colors to critically analyze the ocean scenarios and resources. This area has been explored using different deep learning techniques such as Convolutional neural networks, Residual Networks, and various other models that made use of post-processing techniques for the final enhanced outputs. Hence, this study proposes a novel approach by fine-tuning the Residual-UNet architecture and combining it with an Efficient Channel Attention network in an optimized architecture for the task of underwater image restoration. Channel attention mechanisms have a great ability to reduce model complexity by making use of fewer parameters and increasing the performance of the model. One such mechanism, called Efficient Channel Attention, has been incorporated into the Residual-UNet model to extract features efficiently and obtain an enhanced underwater image that is true to the ground truth image by making use of the HardSwish activation function. The dataset used to train the Residual-UNet model is LSUI (Large Scale Underwater Images), a benchmark dataset consisting of various real underwater images. The proposed model obtained SSIM and PSNR values of 0.9998 and 74.851 dB respectively and outperformed other models by a margin of 4%. To determine the superiority of the Efficient Channel Attention network optimally integrated into the Residual-UNet model, various base models without this attention module have also been trained and evaluated.