A Novel Approach for Underwater Object Detection through Deep Intense-Net for Ocean Conservation Systems

Underwater imaging is a robust tool for hydrographic analysis investigating aqua life possibilities and various research activities. An underwater environment is a unique environment, with frequently varying luminance and objects that differ in appearance compared with the above-water environment. Considering a few challenges, the proposed system is focused on deriving an optimum prediction model, which would differentiate and animate non-animated bodies, which include garbage, debris, etc. The model system uses the Stacked-CNN architecture, which has been optimized and forms a Deep Intense-Net which is customized with a particular focus on underwater objects. In this, the input images are labeled and converted into train images with back annotated bounding boxed features. Image samples of living organisms and non-living things in an underwater environment have been captured. The dataset is formed by combining a few real-time Google images with the brackish dataset. Among these, 75% of the images were used for the training process and the rest 25% was utilized for the testing or validation process. If a new input is forwarded to the network, it will map the features of the input image with the trained underwater images and give its output. These mapped features are combined to create a robust feature box that ensures the prediction quality. The model is being simulated on the MATLAB 2017 platform and the quantitative measures are done based on true positive rate, true negative rate, false-positive rate, and false-negative rate to provide relevant accuracy.