Detection Of Trawling Marks From Digital Images

The environmental and economical impacts of fishery activities have received considerable attentions. During the past few decades, a large number of studies have proved that bottom trawling not only caused the decline of many fish stocks, but also damaged the biomass of non-target species and benthic habitats. Bottom trawl causes a number of direct and indirect changes in the ecosystem, include destruction of benthos, drastic reduction of the physical complexity of the habitat, post-fishing mortality and long-term changes to the benthic community structure. Therefore, it is highly desirable to monitor the damaged level of seabed environment, to understand biological variations associated benthic environment, and to establish necessary management and protection.Nowadays, the progress in underwater roberts, manned and unmanned submersibles, underwater imaging technologies, image processing, pattern recognition, as well as geographic information system, has been used in the research and management of oceanic fishery. It provides efficient ways to monitor the trawling activities in term of area, intensity and durations, and to improve scientist understanding the effects of fishing trawl on benthic habits, resource protection and management. It is helpful in decision-making for the administration. However, as far as know, works on the recognition of trawling marks based on image analysis remain rare, and there is no reported works so far in China. In this dissertation, a method to deal with automatic detection and recognition of trawling marks from video images was presented, and some key problems such as the line detection of trawling marks, the expression and extraction of shape features, using artificial neural network (ANN) for trawling marks classification, were addressed. The strategy of recognizing trawling marks based on geometric features is established.The results of the dissertation are presented as follow: (1) An underwater towed optical vehicle and the images acquisition system was developed, and also an approach for image denoising and enhancement was proposed and tested. (2) A new method for line detection, which was named as the extended Progressive Probabilistic Hough Transform (EPPHT), was proposed to reduce the compute complexity. The advantage of new method is that it makes use of line feature criteria to detect the edge segments and to minimize the poll size. (3) A novel approach based on the stationary wavelet transform for the edge detection and line segment detection, which takes advantage of the properties of locality and multiscale analysis of wavelet transform to eliminate the noise, and to improve the reliable detection and location of the edges. (4) A methodology for geometric shape features extraction in seabed images was presented. The features we extracted were invariant to rotation, translation, intensity, type of seabed and illumination changes, and it is appeared as an appealing alternative to classical texture-based features. (5) A scheme based on neural classifier for trawling marks discrimination was provided. It could be served as a general purpose classification algorithm for seabed image analysis.