DOSDet: A Sequential Feature Enhancement Network for Deck Oil Spill Detection

During refueling process, ships experience deck oil leakage. The existing technology relies on hardware devices for detection. Due to the high viscosity and poor transparency of ship oil, affecting the monitoring effect. We considered the temporal features between images and used image detection methods to detect deck oil spills. Therefore, this paper proposes a DOSDet network to detect deck oil spills from a fixed perspective using temporal features. DOSDet introduces the GRC module and SFFN module, which solves the problem of being unable to input images of any size when processing image sequences, and performs preliminary sequence feature extraction on the input continuous feature maps; The SFFN module deeply explores potential information of each channel and the relationships between channels through extracted sequence features, enhancing the processing ability of channel level important information in input feature map. We also proposed a deck oil leakage dataset to fill the gap in the domain dataset.