Deep Learning Based Algae Detection Method

The ocean is an important part of the ecosystem and is closely related to our lives. Detecting the status of algae in the ocean contributes to the protection of the marine environment. With the continuous development of target detection technology, small target detection tasks are gradually applied to the task of monitoring marine organisms. We use two-stage cascade RCNN with Res2Net, ResNeSt, CBNet, ConvNeXt and DetectoRS backbone. Secondly, data pre-processing was used with blur, motion blur, MixUp, random rotation and other data enhancements. Then the pseudo label training model is used as a pre-training model. And model ensemble is used to improve the inference results. Finally Post-processing is performed using reduced bbox. We conduct extensive experiments on the dataset and achieve the performance of 0.562.