AUTHOR=Zheng Kaijian , Yang Renyou , Li Rifu , Yang Liang , Qin Hao , Li Ziyun 

TITLE=A dual stream hierarchical transformer for starvation grading of golden pomfret in marine aquaculture

JOURNAL=Frontiers in Marine Science

VOLUME=Volume 9 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/marine-science/articles/10.3389/fmars.2022.1039898

DOI=10.3389/fmars.2022.1039898

ISSN=2296-7745

ABSTRACT=Fish starvation grading can provide feeding information for aquaculture, saving the cost of lures and also helping to promote the unmanned and intelligent process of offshore aquaculture. In this paper, we conducted a study on a marine aquaculture vessel for golden pomfret and collected aquaculture data for fish starvation grading with an underwater camera. First, we built a dual stream dataset, consisting of spatial channel and temporal channel, and the fish school starvation status were divided into five levels. Then, according to the marine image characteristics, we proposed a Hierarchical Convolutional Network for feature extraction, and the image feature blocks of different channels were merged by Composite Fusion Convolution and Transformer to obtain the grading results. Finally, we conducted qualitative and quantitative experiments, and the proposed model DSHT achieved the state-of-the-art starvation grading performance with a test accuracy of 98.05%, which is significantly higher than other mainstream models. Meanwhile, we conducted field tests on the vessel to prove that the model can be applied to the mariculture environment of golden pompano.