Monitoring of endangered fish using image processing and AI tools

Abstract

Marine life constitutes half of earth’s total biodiversity. But preservation and monitoring of them efficiently face setbacks largely due to technological limitations and economic reasons. Technological challenges include lack of effective image processing methods curtailed to underwater environment. Underwater image processing, object detection and classification have always been a challenge to accomplish through traditional methods. The methods including sonar radiation produces results, but they are nowhere near economical or accurate as intended. Alternatively, research has been conducted in solid and stationary object detection. Combining the knowledge of the already existing researches done in object detection, in this thesis, we compare performances of various classification algorithms using the data extracted from images of fishes taken in various luminous conditions. For this paper we will only consider large to medium sized fishes but exclude other non-chordate bio-life. The first step is to prepare the images for suitable feature extraction.The next step is to extract suitable features from the available images of the dataset.The next step is to classify the fishes through four different classifiers (SVM, KNN, NB and Random forest classifier) on the basis of the features we have extracted. Lastly we compare the relative performances of these classifiers.