Identification and Detection of Marine Mammal Dorsal Fin Based on Deformable-DETR

The development of human society has significantly impacted the natural habitats of certain marine animals, pushing many marine mammals to the brink of extinction. To protect and observe endangered animal populations, this article proposes a method for identifying marine mammal dorsal fins. The method utilizes a variable convolutional neural network to recognize dorsal fins, enabling the identification and detection of exposed fin marine mammals, the practicality of this method has been validated in this article.The proposed method employs attention mechanisms and feature fusion to learn distinctive features in different regions. Additionally, modifications are made to the loss function and feature extraction methods. These adjustments facilitate better learning of dorsal fin characteristics, ultimately achieving recognition of the white dolphins.By leveraging this advanced technology, we hope to enhance the conservation of endangered marine species, creating better conditions for their survival and reproduction.