Fig. 1. Architecture of ship detection network

Fig. 2. Block of residual connection (Res-Conv2D)

Fig. 3. Ship models and infrared imaging system. (a) Infrared imaging system; (b) flume; (c) ship models; (d) surface ship

Fig. 4. Samples of ship dataset production. (a) Original data; (b) data labeled by LabelMe

Fig. 5. Variation of binary cross-entropy loss in training processing for four networks

Fig. 6. Variation of pixel accuracy in training processing for four networks

Fig. 7. Output results. (a) Input images; (b) output results of manual annotation; (c) output results of BS-Net; (d) output results of BRS-Net; (e) output results of TS-Net; (f) output results of TRS-Net