The interpretation of imagery produced by autonomous underwater vehicle side scanning sonar system used for locating anti-shipping mines is a crucial operation for naval safety. Traditional methods involved the technical operator manually scanning through a large dataset of sonar imagery, which inevitably introduces human biases/errors. Deep learning algorithms have shown potential to identify mine like objects within sonar imagery at a rapid pace with consistent results and with minimal human operator interaction. Increasingly, the Navy are looking to employ autonomous underwater vehicles fitted with side scanning sonar to perform remote sensing in the search for mines. Acquiring the amount of data required to train a deep learning model is an expensive task, therefore this area of interest currently lacks practical application. The purpose of this exercise was to perform a binary classification task to classify images as either containing a mine like object or not. Are deep learning algorithms more efficient at analysing sonar images, are they worth the investment and what are the limitations facing these algorithms today? The Convolution Neural Network (CNN) model built was trained using sonar images that either contained a mine like object or none, the model would then be evaluated on its ability to classify images with mine like objects within a separate dataset. The results were then compared with a real-world experienced operator attempting to manually complete the same task. Within the scope of testing conducted, the CNN tended to be more liberal in its classifications, as it displayed a higher false positive rate in comparison to the human classifier. Overall, the CNN showed good potential for operational use, considering its ability to determine a classification within a fraction of a second. Predictably, the performance of the Convolution Neural Network appeared to improve as the quality and quantity of training images increased.