Optical Small Target Detection Method by Drone Based on Dual-Modal Image Fusion

Kaixuan Chang; Jianhua Huang; Xiyan Sun; Jian Luo; Shitao Bao; Huansheng Huang

doi:10.3788/LOP241283

Laser & Optoelectronics Progress, Volume. 62, Issue 4, 0428001(2025)

Optical Small Target Detection Method by Drone Based on Dual-Modal Image Fusion

Kaixuan Chang^1,2、*, Jianhua Huang^1,2, Xiyan Sun^1,2, Jian Luo^1,2, Shitao Bao^1,2, and Huansheng Huang^1,2

Author Affiliations

¹Schnool of Information and Communicaiton, Guilin University of Electronic Technology, Guilin 541004, Guangxi , China

²Guangxi Key Laboratory of Precision Navigation Technology and Application, Guilin University of Electronic Technology, Guilin 541004, Guangxi , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(20)

Fig. 1. Network structure diagram of YOLOv8

Download full size

Fig. 2. Network structure diagram of FFDN

Download full size

Fig. 3. Network structure diagram of MFDN

Download full size

Fig. 4. Network structure diagram of BFDN

Download full size

Fig. 5. Network structure diagram of BFDN_YOLOv8

Download full size

Fig. 6. Structure of BRA module

Download full size

Fig. 7. Structure of DCFFB module

Download full size

Fig. 8. Visualization results of comparative experiments on fusion detection frameworks in DroneVehicle dataset

Download full size

Fig. 9. Visualization results of comparative experiment on object detection networks in DroneVehicle dataset

Download full size

Fig. 10. Visualization results of comparative experiment on dual-modal image fusion detection method in DroneVehicle dataset

Download full size

Fig. 11. Visualization results of ablation experiments in DroneVehicle dataset

Download full size

Fig. 12. Visualization results of fusion detection framework comparison experiment in LLVIP dataset

Download full size

Fig. 13. Visualization results of comparative experiment of dual-modal image fusion detection method in LLVIP dataset

Download full size

Table 1. Training parameter settings
View table
Table 1. Training parameter settings
Training parameter Value
Epoch 100
Batch_size 16
Image size 640✕640
Initial learning rate 0.001
Optimizer Adam
Pretrained weights ‒

Table 2. Comparative experiments of fusion detection framework in DroneVehicle dataset

View table

Table 2. Comparative experiments of fusion detection framework in DroneVehicle dataset

Model	Index	Value /%						FPS /（frame/s）	Number of parameters	GFLOPS
Model	Index	All	Car	Truck	Bus	Van	Feright car	FPS /（frame/s）	Number of parameters	GFLOPS
YOLOv8_VI	Precision	69.2	89.9	61.5	88.3	55.1	51.4	357.1	3152336	8.8
	Recall	68.8	80.2	63.5	90.8	52.1	47.4
	mAP	70.8	94.5	63.0	94.3	53.2	49.2
YOLOv8_IR	Precision	65.6	83.9	53.4	86.5	52.5	51.8	357.1	3152336	8.8
	Recall	70.4	94.2	65.5	93.5	50.8	48.0
	mAP	69.7	94.1	61.2	93.9	51.1	48.4
Dual-channel YOLOv8	Precision	70.7	85.9	59.8	86.7	58.9	62.1	232.6	6011670	16.2
	Recall	73.0	95.5	71.3	94.1	54.5	49.7
	mAP	73.4	96.6	65.2	94.5	57.7	53.1
FFDN	Precision	75.0	90.6	68.1	88.5	66.2	61.8	303.0	3008187	8.8
	Recall	73.2	95.2	70.5	93.4	55.6	51.3
	mAP	77.6	97.6	73.5	96.4	63.0	57.8
MFDN	Precision	70.2	89.4	59.9	86.9	58.9	55.7	122.0	3007598	8.8
	Recall	73.3	95.8	68.6	92.9	53.7	55.4
	mAP	75.3	97.5	69.2	95.7	57.2	57.1
BFDN	Precision	75.7	91.3	66.6	89.0	64.3	67.3	238.1	6313683	16.4
	Recall	73.5	95.0	69.8	94.3	59.0	49.6
	mAP	78.7	97.8	73.2	96.3	65.6	60.8

Table 3. Comparative experiment of object detection networks in DroneVehicle dataset

View table

Table 3. Comparative experiment of object detection networks in DroneVehicle dataset

Model	Index	Value /%						FPS /（frame/s）	Number of parameters	GFLOPS
Model	Index	All	Car	Truck	Bus	Van	Feright car	FPS /（frame/s）	Number of parameters	GFLOPS
YOLOv8_VI	Precision	69.2	89.9	61.5	88.3	55.1	51.4	357.1	3152336	8.80
	Recall	68.8	80.2	63.5	90.8	52.1	47.4
	mAP	70.8	94.5	63.0	94.3	53.2	49.2
RT-DETR	Precision	69.1	86.9	58.2	78.8	64.8	57.0	196.1	32816351	10.80
	Recall	67.0	94.7	63.3	91.7	46.6	38.7
	mAP	71.2	93.1	65.9	96.4	53.1	47.3
RTMDet	Precision	74.8	92.3	66.8	90.9	59.2	64.5	21.2	52258000	79.96
	Recall	71.0	90.6	69.7	92.1	50.9	51.3
	mAP	77.3	95.5	73.3	93.0	66.5	58.1
TOOD	Precision	66.2	86.5	63.2	84.4	47.7	49.1	12.7	3202800	158.00
	Recall	64.1	88.9	56.5	87.3	39.6	48.3
	mAP	73.6	91.3	62.3	94.7	59.1	60.6
BFDN_YOLOv8	Precision	77.3	91.7	68.0	91.9	67.8	67.0	153.8	6700671	17.10
	Recall	74.3	94.8	70.5	94.3	58.1	53.8
	mAP	80.0	97.8	75.4	96.6	67.0	63.0

Table 4. Comparative experiment of dual-modal image fusion detection methods in DroneVehicle dataset

View table

Table 4. Comparative experiment of dual-modal image fusion detection methods in DroneVehicle dataset

Model	Index	Value /%
Model	Index	All	Car	Truck	Bus	Van	Feright car
YOLOv8_VI	Precision	69.2	89.9	61.5	88.3	55.1	51.4
	Recall	68.8	80.2	63.5	90.8	52.1	47.4
	mAP	70.8	94.5	63.0	94.3	53.2	49.2
YOLOv8_IR	Precision	65.6	83.9	53.4	86.5	52.5	51.8
	Recall	70.4	94.2	65.5	93.5	50.8	48.0
	mAP	69.7	94.1	61.2	93.9	51.1	48.4
SwinFusion	Precision	67.8	87.8	57.4	85.5	55.9	52.2
	Recall	70.4	93.9	64.6	89.0	53.8	50.7
	mAP	71.6	96.2	63.7	93.1	54.4	50.6
MFEIF	Precision	68.4	88.7	61.4	85.9	54.5	51.6
	Recall	72.0	95.2	70.6	92.5	52.5	49.1
	mAP	72.7	97.1	70.2	95.1	54.0	47.0
NestFuse	Precision	68.4	89.3	57.2	88.9	52.9	53.6
	Recall	71.3	95.8	67.0	94.5	53.4	45.6
	mAP	71.0	96.2	62.3	93.0	51.0	52.5
PIAFuison	Precision	67.5	85.8	56.5	85.7	50.3	59.0
	Recall	67.7	92.3	64.2	92.5	47.9	41.6
	mAP	68.4	93.3	57.1	93.6	48.9	49.1
PSFusion	Precision	71.7	89.8	62.7	86.4	62.7	57.0
	Recall	71.7	95.5	67.6	92.3	53.2	50.0
	mAP	74.7	97.5	68.3	94.7	56.6	56.3
TarDAL	Precision	63.6	89.6	52.8	81.6	51.2	42.7
	Recall	68.1	91.4	69.1	85.9	49.6	44.7
	mAP	69.8	96.5	59.7	92.7	50.8	49.3
BFDN_YOLOv8	Precision	77.3	91.7	68.0	91.9	67.8	67.0
	Recall	74.3	94.8	70.5	94.3	58.1	53.8
	mAP	80.0	97.8	75.4	96.6	67.0	63.0

Table 5. Ablation experiments in DroneVehicle dataset

View table

Table 5. Ablation experiments in DroneVehicle dataset

Model	Index	Value /%						FPS /（frame/s）	Number of parameters	GFLOPS
Model	Index	All	Car	Truck	Bus	Van	Feright car	FPS /（frame/s）	Number of parameters	GFLOPS
Dual-channel YOLOv8	Precision	70.7	85.9	59.8	86.7	58.9	62.1	232.6	6011670	16.2
	Recall	73.0	95.5	71.3	94.1	54.5	49.7
	mAP	73.4	96.6	65.2	94.5	57.7	53.1
w/o BRA	Precision	69.1	85.9	59.3	84.5	56.8	58.9	169.5	6846998	18.5
	Recall	73.2	95.8	70.9	92.9	54.9	51.6
	mAP	73.5	96.7	64	94.6	57.3	55.1
w/o DCFFB	Precision	75.6	90.6	66.9	90.9	62.5	67.0	172.4	5864563	14.1
	Recall	75.7	96.0	71.4	93.9	59.9	57.2
	mAP	79.9	97.9	75.1	96.5	65.8	64.2
BFDN_YOLOv8	Precision	77.3	91.7	68.0	91.9	67.8	67.0	153.8	6700671	17.1
	Recall	74.3	94.8	70.5	94.3	58.1	53.8
	mAP	80.0	97.8	75.4	96.6	67.0	63.0

Table 6. Comparative experiment of fusion detection framework in LLVIP dataset
View table
Table 6. Comparative experiment of fusion detection framework in LLVIP dataset
Model Precision /% Recall /% mAP /%
YOLOv8_VI 81.7 67.1 76.0
YOLOv8_IR 86.2 74.0 78.7
Dual-channel YOLOv8 84.1 75.8 84.6
FFDN 88.6 72.8 84.1
MFDN 87.4 76.8 85.9
BFDN 85.1 80.7 87.0

Table 7. Comparative experiment of dual-modal image fusion detection methods in LLVIP dataset

View table

Table 7. Comparative experiment of dual-modal image fusion detection methods in LLVIP dataset

Model	Precision /%	Recall /%	mAP /%
YOLOv8_VI	81.7	67.1	76.0
YOLOv8_IR	86.2	74.0	78.7
SwinFusion	88.2	78.6	87.8
MFEIF	88.2	80.1	87.7
NestFuse	88.6	78.2	88.1
PIAFusion	89.7	79.7	88.6
PSFusion	87.2	75.8	84.1
TarDAL	88.4	79.3	87.9
BFDN_YOLOv8	88.6	80.7	88.7

Tools

Get Citation

Copy Citation Text

Kaixuan Chang, Jianhua Huang, Xiyan Sun, Jian Luo, Shitao Bao, Huansheng Huang. Optical Small Target Detection Method by Drone Based on Dual-Modal Image Fusion[J]. Laser & Optoelectronics Progress, 2025, 62(4): 0428001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Remote Sensing and Sensors

Received: May. 14, 2024

Accepted: Jun. 27, 2024

Published Online: Feb. 18, 2025

The Author Email:

DOI:10.3788/LOP241283

CSTR:32186.14.LOP241283

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Training parameter settings

Table 1. Training parameter settings

Table 2. Comparative experiments of fusion detection framework in DroneVehicle dataset

Table 2. Comparative experiments of fusion detection framework in DroneVehicle dataset

Table 3. Comparative experiment of object detection networks in DroneVehicle dataset

Table 3. Comparative experiment of object detection networks in DroneVehicle dataset

Table 4. Comparative experiment of dual-modal image fusion detection methods in DroneVehicle dataset

Table 4. Comparative experiment of dual-modal image fusion detection methods in DroneVehicle dataset

Table 5. Ablation experiments in DroneVehicle dataset

Table 5. Ablation experiments in DroneVehicle dataset

Table 6. Comparative experiment of fusion detection framework in LLVIP dataset

Table 6. Comparative experiment of fusion detection framework in LLVIP dataset

Table 7. Comparative experiment of dual-modal image fusion detection methods in LLVIP dataset

Table 7. Comparative experiment of dual-modal image fusion detection methods in LLVIP dataset