Ship Detection in Spaceborne Infrared Image Based on Lightweight CNN and Multisource Feature Cascade Decision
Nan Wang,Bo Li,Xingxing Wei,Yonghua Wang,Huanqian Yan
DOI: https://doi.org/10.1109/tgrs.2020.3008993
IF: 8.2
2021-05-01
IEEE Transactions on Geoscience and Remote Sensing
Abstract:Infrared remote-sensing images have irreplaceable value in military and civilian research, such as remote surveillance and military reconnaissance. However, under the conditions of complex scenes, infrared ship detection still faces great challenges. Most importantly, because of the limited hardware resource, the spaceborne satellites usually have weak computational processing ability, which makes the traditional convolution neural network (CNN)-based detection algorithms difficult to show their power. In terms of the above facts, this article proposes a high-performance but low-computation and storage-efficient ship detection algorithm to adapt the severe spaceborne environment. Overall, our method contains the following technical steps: 1) we first present a novel iterative precise segmentation of land and sea algorithm to preprocess the complex and diverse remote-sensing scenes; 2) the multivariate Gaussian distribution is then selected to extract the ship target candidate regions to guarantee the detection recall; 3) we adopt the optical panchromatic data to assist the limited infrared data training; and 4) the cascade decision of multisource features including global and local cues is next utilized to gradually eliminate false alarms. Due to the high efficiency, the proposed method can implement well on the hardware platform of DSP and field-programmable gate array (FPGA) architecture. We conduct a series of experiments and compare with the state-of-the-art object detection algorithms. Experimental results show that our method has fewer parameters but can achieve strong detection robustness against the noise, cloud and reef interfere, which verifies the effectiveness of the proposed method.
imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geochemistry & geophysics