SPEANet: Structural Prior Enhanced Attention Network for Parameter-Efficient Remote Sensing Object Detection

Wei Lu; Junjie Li; Feifei Sang; Si-Bao Chen

cs.CVAug 15, 2026

SPEANet: Structural Prior Enhanced Attention Network for Parameter-Efficient Remote Sensing Object Detection

Authors: Wei Lu, Junjie Li, Feifei Sang, Si-Bao Chen

Organizations: MOE Key Lab of ICSP, IMIS Lab of Anhui, Anhui Provincial Key Lab of Multimodal Cognitive Computation, Zenmorn-AHU · AI Joint Lab, School of Computer Science and Technology, Anhui University, Hefei 230601, China

Abstract

Remote sensing object detection (RSOD) requires compact backbones capable of preserving weak geometric cues under extreme scale variation and background clutter. Fixed structural operators provide complementary contour and frequency responses without introducing learnable operator coefficients. However, directly injecting these responses can amplify content-irrelevant textures, while applying a uniform operator design across the hierarchy may be poorly matched to stage-specific representation requirements. We propose the Structural Prior Enhanced Attention Network (SPEANet), a parameter-efficient RSOD backbone that integrates fixed operators through stage-specific prior extraction and context-conditioned response modulation. SPEANet assigns smoothed contour and multi-order directional modeling to shallow, high-resolution features, while employing a compact approximation-detail interaction mechanism in deeper stages. Learned spatial gates regulate the resulting prior responses before residual fusion. Experiments on five benchmarks, together with evaluations across seven detection frameworks on DOTA-v1.0, achieve a favorable accuracy-parameter trade-off. With Oriented R-CNN, SPEANet achieves 78.55% mAP on DOTA-v1.0, 72.24% mAP on DOTA-v1.5, and 67.30% mAP on DIOR-R using 23.0M total parameters, including a 5.97M-parameter backbone.

Figures & tables

Panel of Figure 1 — Figure 1: Top : Accuracy-parameter comparison on the DOTA-v1.0 test set ( 33 ) . Bottom : Representative detections in scenes containing weak object structures and background clutter. Red, yellow, and green denote false positives, false negatives, and true positives, respectively.

Figure 2: Overview of SPEANet. (a) Four-stage hierarchical backbone with stage-specific operator assignment. (b) HPRB, which combines prior extraction, context-conditioned modulation, and residual encoding. (c) MDP-FASE for multi-order directional modeling at Stage I. (d) WASM for approximation-detail interaction at Stages II-IV.

Figure 3: Context-Guided Frequency Filtering (CGFF) module. A dilated context encoder predicts a single-channel spatial gate that modulates the fixed-operator response.

Figure 4: Frequency-Aware Scaling and Encoding (FASE) module. A globally aggregated descriptor conditions the scaling of the pointwise-transformed feature branch.

One Stage
Method	Backbone	#P $\downarrow$	mAP $\uparrow$
R 3 Det ( 38 )	ResNet-50 ( 14 )	41.9M	69.70
SASM ( 15 )	ResNet-50 ( 14 )	36.6M	74.92
O-RepPoints ( 16 )	ResNet-50 ( 14 )	36.6M	75.97
R 3 Det-GWD ( 39 )	ResNet-50 ( 14 )	41.9M	76.34
R 3 Det-KLD ( 42 )	ResNet-50 ( 14 )	41.9M	77.36

Table 1: Performance comparison on the DOTA-v1.0 test set under single-scale training and testing. The best results are shown in bold, and the second-best results are underlined.

Framework		ResNet-50	PKINet-S	SPEANet
Framework		(23.28M)	(13.70M)	(5.97M)
One Stage	R-FCOS ( 29 )	72.45	74.86	76.27
	R 3 Det ( 38 )	69.70	75.89	76.32
	S 2 ANet ( 10 )	74.12	77.83	78.16
Two Stage	FR-O ( 27 )	73.17	76.45	76.86
	Roi Trans. ( 7 )	74.05	77.17	79.15

Table 2: Comparison of mAP across different detectors and backbones (ResNet-50 and PKINet-S) on the DOTA-v1.0 test set. Only backbone parameters are included.

Method	#P $\downarrow$	LV	SH	BC	HC	mAP $\uparrow$
RetinaNet-O ( 18 )	36.4M	56.79	73.31	76.02	48.06	59.16
FR-O ( 27 )	41.1M	68.98	79.37	77.38	60.47	62.00
Mask R-CNN ( 13 )	44.4M	71.34	79.75	74.21	57.81	62.67
HTC ( 2 )	77.5M	73.31	80.31	75.12	55.87	63.40
ReDet ( 11 )	31.6M	75.73	80.92	75.81	63.33	66.86
LSKNet-S ( 17 )	31.0M	77.45	81.17	79.44	55.81	70.26

Table 3: Performance comparison on the DOTA-v1.5 test set. Representative categories involving small objects, elongated structures, or cluttered backgrounds are reported; complete category results are provided in Supplementary Material.

Method	Backbone	Overall / Backbone		mAP (%) $\uparrow$
Method	Backbone	#P $\downarrow$ (M)	FLOPs(G)	mAP (%) $\uparrow$
FR-O ( 27 )	ResNet50	41.1 / 23.3	148.2 / 52.4	59.54
G.V. ( 36 )	ResNet50	41.1 / 23.3	148.2 / 52.4	60.06
Roi Trans. ( 7 )	ResNet50	55.1 / 23.3	176.2 / 52.4	63.87
LSKNet ( 17 )	LSKNet-S	31.0 / 13.8	125.1 / 32.9	65.90
Oriented Rep ( 16 )	ResNet50	36.6 / 23.3	118.7 / 52.4	66.71

Table 4: Experimental results on the DIOR-R test set. #P and FLOPs were measured at a

800\times 800

resolution.

Method	AP 50:95	AP 50	AP 75	#P $\downarrow$	FLOPs
Rotated Faster RCNN ( 27 )	32.5	70.1	24.3	41.1M	292G
Rotated RetinaNet ( 18 )	26.8	63.4	16.2	36.2M	800G
RoI Trans. ( 7 )	36.0	73.0	30.1	55.1M	306G
Gliding Vertex ( 36 )	31.7	70.8	22.6	41.1M	292G
Oriented RCNN ( 34 )	34.4	70.7	28.6	41.1M	292G
S 2 A-Net ( 10 )	28.3	69.6	13.1	38.6M	733G

Table 5: Comparison on the SODA-A test set. #P and FLOPs are measured at

1200\times 1200

resolution.

Method	AP 50	AP 75	mAP	#P $\downarrow$	FLOPs
R 3 Det ( 38 )	60.2	36.7	34.8	42.0M	362.2G
S 2 ANet ( 10 )	63.1	38.7	36.7	38.8M	216.5G
R-FCOS ( 29 )	63.5	41.8	39.1	32.2M	226.9G
RoI Trans. ( 7 )	65.4	43.4	40.7	55.3M	244.7G
StripRCNN ( 43 )	66.1	44.0	41.0	45.2M	231.8G
CFA ( 9 )	66.2	43.2	40.6	36.9M	213.6G

Table 6: Comparison with representative methods on the MODA test set. FLOPs are measured at an input resolution of

1200\times 900

; #P denotes the total number of parameters.

Figure 5: Detection results and CGFF gating maps on the DOTA-v1.0 test set. (a) PKINet-S. (b) SPEANet. (c) Spatial gating maps produced by CGFF.

CLoG-Stem	MDP-FASE	WASM	#P $\downarrow$	mAP $\uparrow$	Category AP
CLoG-Stem	MDP-FASE	WASM	#P $\downarrow$	mAP $\uparrow$	SV	LV	SP
✓			5.351M	76.581	78.75	84.51	72.81
	✓		6.526M	77.184	78.43	85.60	73.23
		✓	4.786M	77.605	78.53	84.53	78.62
✓	✓		6.531M	78.054	79.38	85.15	80.24
✓		✓	4.791M	77.778	78.86	84.92	78.99

Table 7: Ablation study on DOTA-v1.0. Variants containing only MDP-FASE or WASM apply the corresponding module to all backbone stages. When both are enabled, MDP-FASE is used in Stage I and WASM in Stages II–IV. All variants use 100-epoch ImageNet-1K pretraining and Oriented R-CNN. #P denotes backbone parameters. SV, LV, and SP denote Small Vehicle, Large Vehicle, and Swimming Pool. Complete results are provided in Supplementary Material.

Appendix figures & tables9 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 6: Detailed overview of the SPEANet architecture

$\{C_{1},C_{2},C_{3},C_{4}\}$	$\{N_{1},N_{2},N_{3},N_{4}\}$	Down. Ratio	Configuration
$\{32,64,128,256\}$	$\{1,4,4,2\}$	$1/4,1/8,1/16,1/32$	{M W W W}
Params and FLOPs $\dagger$ for each stage			Overall
0.068M, 0.465M, 1.806M, 3.632M			5.971M
7.253G, 7.702G, 7.441G, 3.744G			26.14G

Appendix

Table 8: Detailed configurations of SPEANet.

C_{i}

means the number of output channels in each stage and

N_{i}

means the number of HPRBs in each stage. Configuration denotes the strategy adopted at each stage, while ‘M’ and ‘W’ stand for MDP-FASE and WASM. Both the parameters and FLOPs reported for each stage include those of the stage itself and its associated downsampling module. Overall only represents the backbone network, and

\dagger

indicates that Params and FLOPs are measured using an input

\mathbf{I}\in\mathbb{R}^{1024\times 1024\times 3}

Dataset Name	Images	Classes	Instances	Resolution
DOTA-v1.0	2,806	15	188,282	$\mathrm{800}^{2}$ - $\mathrm{20,000}^{2}$
DOTA-v1.5	2,806	16	403,318	$\mathrm{800}^{2}$ - $\mathrm{20,000}^{2}$
DIOR-R	23,463	20	192,472	800 $\times$ 800
SODA-A	2,513	9	872,069	4,761 $\times$ 2,777 †
MODA	14,041	8	330,191	1,200 $\times$ 900

Appendix

Table 9: Summary of five benchmark datasets. † represents average resolution.

Method	Backbone	#P $\downarrow$	PL	BD	BR	GTF	SV	LV	SH	TC	BC	ST	SBF	RA	HA	SP	HC	mAP $\uparrow$
One Stage
R 3 Det ( 38 )	ResNet-50 ( 14 )	41.9M	89.00	75.60	46.64	67.09	76.18	73.40	79.02	90.88	78.62	84.88	59.00	61.16	63.65	62.39	37.94	69.70
SASM ( 15 )	ResNet-50 ( 14 )	36.6M	86.42	78.97	52.47	69.84	77.30	75.99	86.72	90.89	82.63	85.66	60.13	68.25	73.98	72.22	62.37	74.92
O-RepPoints ( 16 )	ResNet-50 ( 14 )	36.6M	87.02	83.17	54.13	71.16	80.18	78.40	87.28	90.90	85.97	86.25	59.90	70.49	73.53	72.27	58.97	75.97
R 3 Det-GWD ( 39 )	ResNet-50 ( 14 )	41.9M	88.82	82.94	55.63	72.75	78.52	83.10	87.46	90.21	86.36	85.44	64.70	61.41	73.46	76.94	57.38	76.34
R 3 Det-KLD ( 42 )	ResNet-50 ( 14 )	41.9M	88.90	84.17	55.80	69.35	78.72	84.08	87.00	89.75	84.32	85.73	64.74	61.80	76.62	78.49	70.89	77.36

Appendix

Table 10: Performance comparison on the DOTA-v1.0 test set. The best results are shown in bold, and the second-best results are underlined. Results for competing methods follow the corresponding published reports ( 20 ) .

Method	#P $\downarrow$	PL	BD	BR	GTF	SV	LV	SH	TC	BC	ST	SBF	RA	HA	SP	HC	CC	mAP $\uparrow$
RetinaNet-O ( 18 )	36.4M	71.43	77.64	42.12	64.65	44.53	56.79	73.31	90.84	76.02	59.96	46.95	69.24	59.65	64.52	48.06	0.83	59.16
FR-O ( 27 )	41.1M	71.89	74.47	44.45	59.87	51.28	68.98	79.37	90.78	77.38	67.50	47.75	69.72	61.22	65.28	60.47	1.54	62.00
Mask R-CNN ( 13 )	44.4M	76.84	73.51	49.90	57.80	51.31	71.34	79.75	90.46	74.21	66.07	46.21	70.61	63.07	64.46	57.81	9.42	62.67
HTC ( 2 )	77.5M	77.80	73.67	51.40	63.99	51.54	73.31	80.31	90.48	75.12	67.34	48.51	70.63	64.84	64.48	55.87	5.15	63.40
ReDet ( 11 )	31.6M	79.20	82.81	51.92	71.41	52.38	75.73	80.92	90.83	75.81	68.64	49.29	72.03	73.36	70.55	63.33	11.53	66.86
LSKNet-S ( 17 )	31.0M	72.05	84.94	55.41	74.93	52.42	77.45	81.17	90.85	79.44	69.00	62.10	73.72	77.49	75.29	55.81	42.19	70.26

Appendix

Table 11: Performance comparison on DOTA-v1.5 test set. The best results are shown in bold and the second-best results are underlined. Results for competing methods follow the corresponding published reports ( 20 ; 31 ) .

CLoG-Stem	MDP-FASE	WASM	#P $\downarrow$	PL	BD	BR	GTF	SV	LV	SH	TC	BC	ST	SBF	RA	HA	SP	HC	mAP $\uparrow$
✓			5.351M	89.23	82.10	54.15	73.79	78.75	84.51	87.89	90.73	87.09	85.94	61.44	62.98	75.60	72.81	61.70	76.581
	✓		6.526M	88.85	82.19	55.88	72.27	78.43	85.60	88.30	90.67	85.70	85.23	63.70	65.03	76.83	73.23	65.86	77.184
		✓	4.786M	88.97	83.27	53.57	76.69	78.53	84.53	87.88	90.62	84.77	84.40	63.92	65.12	76.98	78.62	66.20	77.605
✓	✓		6.531M	89.02	83.10	55.23	73.49	79.38	85.15	88.36	90.73	87.15	85.06	66.90	66.86	76.84	80.24	63.30	78.054
✓		✓	4.791M	88.90	83.47	54.92	75.76	78.86	84.92	88.17	90.72	87.40	84.75	63.25	63.33	76.92	78.99	66.29	77.778
	✓	✓	5.966M	89.07	82.94	55.55	73.59	79.96	85.51	88.16	90.55	87.98	86.22	64.86	66.85	77.09	73.52	68.47	78.022

Appendix

Table 12: Ablation of SPEANet on DOTA-v1.0. All variants use 100-epoch ImageNet-1K pretraining and the Oriented R-CNN detector. #P denotes backbone parameters.

Figure 7: Visual comparison of different SPEANet variants in the ablation study. “Overall” represents the complete SPEANet model while “w/o” denotes the removal of the corresponding component.

Panel of Figure 8 — Figure 8: Qualitative visualization results of SPEANet. Top: Visualization of detection results and CGFF gating maps on different datasets. The left and right examples are from DIOR-R and SODA-A, respectively. Bottom: Visualization of detection results compared with Roi Trans. ( 7 ) on MODA dataset.

Figure 9: Comparison of the detection results obtained by different methods on the DOTA-v1.5 test set.

Explore similar work

PKINet-v2: Towards Powerful and Efficient Poly-Kernel Remote Sensing Object Detection
Mar 17, 2026Xinhao Cai, Liulei Li, Gensheng Pei +3Remote Sensing Efficient Object Detection
Nanjing University of Science and Technology · State Key Lab. of Intelligent Manufacturing of Advanced Construction Machinery · Zhejiang University +1
RDNet: Region Proportion-Aware Dynamic Adaptive Salient Object Detection Network in Optical Remote Sensing Images
Mar 12, 2026Bin Wan, Runmin Cong, Xiaofei Zhou +3Salient Object Detection Multi-Scale Attention
School of Control Science and Engineering, Shandong University, Jinan 250061, China · Key Laboratory of Machine Intelligence and System Control, Ministry of Education, Jinan 250061, China · School of Automation, Hangzhou Dianzi University, Hangzhou 310018, China +2
BMCR: Adaptive Backbone Module Composition via Reinforcement Learning for Remote Sensing Object Detection
Jun 4, 2026Wenlin Liu, Xikun Hu, Ping ZhongReinforcement Learning Dynamic Neural Networks
College of Electronic Science and Technology, National University of Defense Technology, Changsha 410073, China

Mar 17, 2026J/K move · Enter open · S save

PKINet-v2: Towards Powerful and Efficient Poly-Kernel Remote Sensing Object Detection

Xinhao Cai, Liulei Li, Gensheng Pei, Zeren Sun+2

Nanjing University of Science and Technology · State Key Lab. of Intelligent Manufacturing of Advanced Construction Machinery · Zhejiang University+1

Remote Sensing Efficient Object Detection Convolutional Neural Networks Remote Sensing Small Object Detection

Open paper AskPDF arXiv

Object detection in remote sensing images (RSIs) is challenged by the coexistence of geometric and spatial complexity: targets may appear with diverse aspect ratios, while spanning a wide range of object sizes under varied contexts. Existing RSI backbones address the two challenges separately, either by adopting anisotropic strip kernels to model slender targets or by using isotropic large kernels to capture broader context. However, such isolated treatments lead to complementary drawbacks: the strip-only design can disrupt spatial coherence for regular-shaped objects and weaken tiny details, whereas isotropic large kernels often introduce severe background noise and geometric mismatch for slender structures. In this paper, we extend PKINet, and present a powerful and efficient backbone that jointly handles both challenges within a unified paradigm named Poly Kernel Inception Network v2 (PKINet-v2). PKINet-v2 synergizes anisotropic axial-strip convolutions with isotropic square kernels and builds a multi-scope receptive field, preserving fine-grained local textures while progressively aggregating long-range context across scales. To enable efficient deployment, we further introduce a Heterogeneous Kernel Re-parameterization (HKR) Strategy that fuses all heterogeneous branches into a single depth-wise convolution for inference, eliminating fragmented kernel launches without accuracy loss. Extensive experiments on four widely-used benchmarks, including DOTA-v1.0, DOTA-v1.5, HRSC2016, and DIOR-R, demonstrate that PKINet-v2 achieves state-of-the-art accuracy while delivering a

\textbf{3.9}\times

FPS acceleration compared to PKINet-v1, surpassing previous remote sensing backbones in both effectiveness and efficiency.

Abstract

Explore similar work

PKINet-v2: Towards Powerful and Efficient Poly-Kernel Remote Sensing Object Detection

RDNet: Region Proportion-Aware Dynamic Adaptive Salient Object Detection Network in Optical Remote Sensing Images

BMCR: Adaptive Backbone Module Composition via Reinforcement Learning for Remote Sensing Object Detection