Segmentation method for overlapping fish images based on point detection and pairing

Lu Zhang,Wensheng Li,Chunhong Liu,Qingling Duan
DOI: https://doi.org/10.1007/s10499-022-00904-9
2022-06-05
Aquaculture International
Abstract:It is important to segment fish images accurately for realizing automatic and welfare fish farming. In aquaculture, there is a large amount of overlap between fish bodies due to the three-dimensional water space in which they dwell and various swimming gestures. The majority of existing fish image segmentation methods are for non-overlapping fish and aren't applied to actual fish segmentation. To address the above problems, a segmentation method based on point search and pairing for overlapping fish images was proposed, using crucian carp as the research object. The top view images of crucian carp in the fish tank were collected and pre-processed. The overlapping crucian carp target was recognized based on the morphological factors and segmented using the proposed point detection and pairing method, which can automatically determine the number of segmentation lines according to the shape of the overlapping crucian carp. The proposed method outperformed the comparison methods on a real crucian carp image dataset, with a segmentation precision of 98.86%, segmentation recall of 97.52%, mean absolute error of 0.1180, average intersection over union of 0.8311, average precision of 90.63%, and average recall of 91.52%, demonstrating that it does well in segmenting overlapping crucian carp images. The proposed method for overlapping fish image is of great significance for well completing the basic farming operations, promoting the welfare farming of fish, and improving the benefits of aquaculture enterprises.
fisheries
What problem does this paper attempt to address?