HRRSODN: high-Resolution remote sensing object detection network fusion of road-adjacent relationships

Figures & data

Figure 1. Network structure diagram of HRRSODN.

Figure 2. Road binary map.

Figure 3. Coordinate attention flow chart for fusing spatial relationships.

Figure 4. POI points in China.

Figure 5. Road network in China.

Figure 6. POI point distribution of gas stations in China.

Table 1. Ablation experiments of YOLOV7 introduced into HRRSODN.

Dataset	Method	AP (%)				F1 Score
		Toll station	Service area	Gas station	mAP (IOU = 0.5)
SR	YOLOV7	95.7	96.0	81.4	91.1	88.0
	YOLOV7 + HRRSODN-CA	96.1	95.1	85.6	92.3	89.0
	YOLOV7+ HRRSODN-Rloss	94.4	96.6	84.6	91.9	89.0
	YOLOV7+ HRRSODN-CA + HRRSODN-Rloss	96.3	96.0	86.2	92.9	90.0

Table 2. Ablation experiments of YOLOV5 introduced into HRRSODN.

Dataset	Method	AP (%)				F1 Score
		Toll station	Service area	Gas station	mAP (IOU = 0.5)
SR	YOLOV5	93.0	94.3	79.4	88.9	88.5
	YOLOV5 + HRRSODN-CA	93.6	94.8	80.1	89.5	89.0
	YOLOV5 + HRRSODN-Rloss	93.1	95.2	82.2	90.2	89.0
	YOLOV5 + HRRSODN-CA + HRRSODN-Rloss	94.4	95.3	84.8	91.5	90.0

Table 3. Ablation experiments of Centernet introduced into HRRSODN.

Dataset	Method	AP (%)				F1 Score
		Toll station	Service area	Gas station	mAP (IOU = 0.5)
SR	Centernet	86.0	90.0	72.0	82.4	80.0
	Centernet + HRRSODN-CA	92.5	92.3	79.2	88.0	82.7
	Centernet + HRRSODN-Rloss	92.3	94.1	74.9	87.1	83.5
	Centernet + HRRSODN-CA + HRRSODN-Rloss	94.0	94.9	78.7	89.2	88.0

Table 4. Ablation experiments of HRRSODN on the gas station dataset.

Dataset	Method	AP (%)	F1
Gas station	YOLOV7	0.698	0.69
	YOLOV7+ HRRSODN-CA	0.712	0.72
	YOLOV7+ HRRSODN-Rloss	0.713	0.70
	YOLOV7+ HRRSODN-CA + HRRSODN-Rloss	0.751	0.72

Figure 7. Detection effect of HRRSODN.

Figure 8. Prediction performance of each network.

Table 5. Comparative experiments of HRRSODN on the SR dataset.

Dataset	Method	AP (%)				F1 Score
		Toll station	Service area	Gas station	mAP (IOU = 0.5)
SR	Faster R-CNN	91.2	95.6	69.5	84.3	66.0
	Centernet	86.0	90.0	72.0	82.7	80.0
	YOLOV5	93.0	94.3	79.4	88.9	88.0
	YOLOV6	91.3	95.4	81.5	89.4	88.5
	Deformable-DETR	95.3	95.5	85.5	92.1	89.4
	Proposed	96.3	96.0	86.2	92.9	90.0

Table 6. Comparative experiments of HRRSODN on the DIOR dataset.

Dataset	Method	AP (%)			F1 Score
		Toll station	Service area	mAP (IOU = 0.5)
DIOR	Faster R-CNN	58.8	82.5	70.7	58.0
	Centernet	79.9	94.2	87.0	83.0
	YOLOV5	85.8	93.8	89.8	89.0
	YOLOV6	82.0	95.2	88.6	88.0
	Deformable-DETR	84.5	96.4	90.5	89.8
	proposed	90.7	98.0	94.4	93.0

Figure 9. Heatmap comparison of multiple networks.

Data availability statement

The paper provides the database used in the current study at baiduy-un(https://pan.baidu.com/s/1FZGTbIPzrjXe1b4qstCD1Q?pwd=xwo4; extraction code: xwo4, accessed on 20 August 2023) and the python code available online at GitHub (https://github.com/suyishao99/HRRSODN, accessed on 20 August 2023).