Surface unmanned vessels often operate in complex environments and require the ability to autonomously plan routes offline. Existing technologies rely heavily on long-range target recognition, while close range optical recognition can effectively serve as a perceptual tool to compensate for deficiencies. Although the marine environment is constantly changing, common target objects for recognition include reefs (on the water surface), islands, offshore production equipment (such as offshore wind turbines), and various types of ships. The images collected from marine environments have the characteristic of grayscale gradient. This article conducts theoretical simulation experiments on the effects and computation time of slant transform, Haar transform, slant Haar transform, Walsh Hadamard transform, Discrete Cosine Transform (DCT), and commonly used differential operators in extracting edges of common targets on water. Based on accuracy and noise resistance, Slant Haar transform was selected as the edge extraction operator to find the image attention area. Then use Hu invariant moments to identify the characteristics of ships in different driving states, providing a new technical approach for edge detection and recognition of common targets on water surfaces.