Underwater Image Sea Cucumber Target Detection Based On Salient Point Features

Nowadays,the development of target detection technology is very mature,but for underwater sea cucumber target detection,due to the dark light and complicated environment,there is no universally applicable method.The traditional image segmentation technology is difficult to extract the sea cucumber outline,and it is difficult to use the supervised learning method to describe the sea cucumber feature only by using features such as color and texture.In this thesis,the SIFT feature is used as the salient point features,and the salient points of the sea cucumber category are screened based on the DOG pyramid layering treatment.Using these salient points to combine with the texture features of the image for secondary screening,the salient points of the sea cucumber category after the secondary screening were obtained.Then use the final sea cucumber category salient points to combine with the super pixel blocks and frame the target to complete the final target detection.The specific research work includes the following three aspects:This thesis proposes the use of SIFT feature points as the salient point features of sea cucumber target detection for the determination of sea cucumber category,and find through the experiment that there are more SIFT feature points in the last one layer and the penultimate layer of the image DOG pyramid,and the generation information is different in the each layer,so use the two layers of SIFT feature points combine with the classifier according to pre-marked sea cucumber positions in the sample image to obtain two sets of models respectively.Using these two sets of models,the two layers of SIFT feature points extracted from the test images were screened for the sea cucumber category,and the salient points obtained by the two-layer screening were combined to obtain the salient points of the sea cucumber category.For the salient points of the sea cucumber category obtained by the layered processing,the texture features are combined to perform the secondary screening of the salient points of the sea cucumber category.Firstly,the sample image is super-pixel segmented,and then the super-pixel blocks containing the salient points of the sea cucumber category are reserved,but in these super-pixel blocks,in addition to the sea cucumber region and the non-sea cucumber region,The GLCM texture features are extracted from these super-pixel blocks and according to pre-marked sea cucumber positions training classifier is used to obtain the model.Then the GLCM texture features are extracted from the super pixel blocks containing the salient points of the test image and the salient points of the sea cucumber category are completed secondary screening by the model.Through experimental analysis,the effectiveness of the proposed method for underwater image sea cucumber target detection is verified.Comparing several methods used in this thesis,it is found that the salient points obtained by the layered processing and combine with the texture features to secondary screening for the sea cucumber target detection works well,and the F1-Measure value reaches 84.0%,this method effectively improves the detection rate.