The underwater sonar image segmentation has been a topic of research for decades. Underwater sonar image is based on the interaction by the echo signal of sound toward the underwater objects or targets. Because of the serious noises polution and the dim target edge, the contrast and resolution of sonar images are obtaind in a decreased quanlity. This paper proposes an improved snake model that focuses on solving underwater target detection and recognition. According to the traditional snake model, it is defined as an energy minimizing spline which is influenced by external constraint forces, and it can guide the image forces to pull toward features, such as lines or edges. Compared with the traditional snake model, this snake model greedy algorithm can converge to the contours more quickly and more stably, especially in complex underwater environments. Examination of the results shows that using snake model greedy algorithm has a more clear shape accuracy.
A Post-processing Fusion Framework for Deep Learning Models for Crop Disease Detection
