Underwater image quality assessment before object detection is a very emerging trend in image processing. In the past several years, a batch of new algorithms for improving underwater images have been presented. But these algorithms are mostly assessed on different datasets or a few picked photographs from the current world. However, it's hard to say how well these algorithms will perform on real-world images or how we can monitor their development. The first complete perceptual investigation and analysis of submarine images Object detection is another important task of underwater images. Due to noise images or lower light intensity, it's hard to detect such objects in an accurate manner. In this paper, we propose using deep learning techniques to improve the quality of underwater images and detect objects. The data contains some noisy data. By using data pre-processing and normalisation techniques, filtration has been done. The balanced data feeds to CNN and executes the convolutional and pooling layers for the extraction of features. Finally, the dense layer classifies the entire data set based on the trained module. The VGG-16 and RESNET-101 deep learning frameworks have been used for classification. Extensive testing has shown that CNN with RESENT gives 96.50% accuracy, while VGG-16 CNN gives 95.80% accuracy on the sonar dataset

image processing, Underwater data analytics, sonar signal analysis, supervised classification, machine learning, deep learning