A lightweight CNN for multi-source infrared ship detection from unmanned marine vehicles

• DOI: 10.1016/j.heliyon.2024.e26229
• Publication Date: 2024-02-13T17:25:23Z
• Authors: Liqian Wang; Yakui Dong; Cheng Fei; Junliang Liu; Shuzhen Fan; Yunxia Liu; Yongfu Li; Zhaojun Liu; Xian Zhao
• Keywords: H1-99; Science (General); 02 engineering and technology; Autonomous ships; Convolutional Neural Network (CNN); Social sciences (General); Q1-390; Maritime surveillance; Infrared ship target; 0202 electrical engineering, electronic engineering, information engineering; Ship detection; Research Article

Infrared ship detection is of great significance due to its broad applicability in maritime surveillance, traffic safety and security. Multiple infrared sensors with different spectral sensitivity provide enhanced sensing capabilities, facilitating complex environments. Nevertheless, current researches lack discussion exploration imagers ranges for marine objects detection. Furthermore, unmanned vehicles (UMVs), e.g., surface (USVs) (USs), perception are usually performed embedded devices limited memory computation resource, which makes traditional convolutional neural network (CNN)-based methods struggle leverage their advantages. Aimed at the task sea object on USVs, this paper provides lightweight CNNs high inference speed that can be deployed devices. It also discusses advantages disadvantages using detection, providing a reference decision-making modules USVs. The proposed method detect ships short-wave (SWIR), long-wave (LWIR) fused images high-performance high-inference an device. Specifically, backbone built from bottleneck depth-separable convolution residuals. Generating redundant feature maps by cheap linear operation neck head networks. learning representation capacities promoted introducing channel spatial attention, redesigning sizes anchor boxes. Comparative experiments conducted dataset we have released contains SWIR, LWIR images. results indicate achieve accuracy but fewer parameters, nearly 60 frames per second (FPS)