Video-Based Hierarchical Species Classification for Longline Fishing Monitoring

The goal of electronic monitoring (EM) longline fishing is to monitor the fish catching activities on vessels, either for regulatory compliance or catch counting. Hierarchical classification based videos allows inexpensive and efficient species identification catches from fishing, where fishes are under severe deformation self-occlusion during process. More importantly, flexibility hierarchical mitigates laborious efforts human reviews by providing confidence scores in different levels. Some related works use cascaded models make predictions per image predict one overlapping data structure dataset advance. However, with a known non-overlapping provided fisheries scientists, our method enforces introduces an training inference strategy video-based data. Our experiments show that proposed outperforms classic flat system significantly ablation study justifies contributions CNN model design, strategy, schemes task.