基于多尺度特征融合的轻量型野外蝙蝠检测

doi:10.11996/JG.j.2095-302X.2025010070

图学学报 ›› 2025, Vol. 46 ›› Issue (1): 70-80.DOI: 10.11996/JG.j.2095-302X.2025010070

• 图像处理与计算机视觉 • 上一篇下一篇

基于多尺度特征融合的轻量型野外蝙蝠检测

王杨¹(), 马唱¹, 胡明¹, 孙涛², 饶元³, 袁振羽¹

1.安徽师范大学计算机与信息学院，安徽芜湖 241002
2.哈尔滨工业大学机电工程学院，黑龙江哈尔滨 150001
3.安徽农业大学信息与计算机学院，安徽合肥 230036

收稿日期:2024-07-25 接受日期:2024-10-14 出版日期:2025-02-28 发布日期:2025-02-14
第一作者:王杨(1971-)，男，教授，博士。主要研究方向为人工智能、增强现实和生态信息学等。E-mail：wycap@126.com
基金资助:
国家自然科学基金(61871412);安徽省自然科学基金重点项目(KJ2019A0938);安徽省自然科学基金重点项目(KJ2021A1314);安徽省自然科学基金重点项目(KJ2019A0979);机器视觉检测安徽省重点实验室开放基金资助(KLMVI-2023-HIT-11);农业农村部农业传感器重点实验室开放课题(KLAS2023KF001);安徽高校自然科学重点研究项目(2022AH052899)

Lightweight wild bat detection method based on multi-scale feature fusion

WANG Yang¹(), MA Chang¹, HU Ming¹, SUN Tao², RAO Yuan³, YUAN Zhenyu¹

1. School of Computer and Information, Anhui Normal University, Wuhu Anhui 241002, China
2. School of Mechanical and Electrical Engineering, Harbin Institute of Technology, Harbin Heilongjiang 150001, China
3. College of Information and Computer Science, Anhui Agricultural University, Hefei Anhui 230036, China

Received:2024-07-25 Accepted:2024-10-14 Published:2025-02-28 Online:2025-02-14
First author：WANG Yang (1971-), professor, Ph.D. His main research interests cover artificial intelligence, augmented reality and ecological informatics, etc. E-mail：wycap@126.com
Supported by:
National Natural Science Foundation of China(61871412);Anhui Provincial Natural Science Foundation Key Project(KJ2019A0938);Anhui Provincial Natural Science Foundation Key Project(KJ2021A1314);Anhui Provincial Natural Science Foundation Key Project(KJ2019A0979);Open Research Fund of Anhui Province Key Laboratory of Machine Vision Inspection(KLMVI-2023-HIT-11);Open Project of the Key Laboratory of Agricultural Sensors, Ministry of Agriculture and Rural Affairs(KLAS2023KF001);Key Research Project of Natural Science in Anhui Universities(2022AH052899)

摘要/Abstract

摘要：

野外蝙蝠检测对于生态保护和科学研究具有重要意义。针对计算资源有限和复杂野外环境带来的挑战，提出了一种轻量型蝙蝠检测模型(LiteDETR-Bat)，旨在实现高效的实时检测。首先为了解决特征映射冗余问题，引入重参数卷积高效层聚合网络(RCELAN)替换传统的ResNet主干网络，采用多分支特征聚合机制，有效降低了计算复杂度和参数量。其次设计了动态采样多尺度特征融合(DS-MFF)，该结构集成空洞卷积和动态采样算子，通过扩大感受野并自适应调整采样位置，优化多尺度特征融合，提升多样化特征处理时模型的灵活性和鲁棒性。最后在安徽省野外环境下采集了一个涵盖多种光照条件、视角变化及蝙蝠形态变化的蝙蝠数据集，并进行了模型性能等相关实验。实验结果表明，该LiteDETR-Bat模型不仅能够使参数量降低了46.5%，mAP达到97.2%，同时在准确性和实时性上相比于YOLO系列算法均取得了一定地提升。LiteDETR-Bat模型为野外蝙蝠的监测与保护工作提供了有力的技术支持，展现了其在生态监测和生物多样性保护中的应用潜力。

关键词: 野外蝙蝠, RT-DETR, 多尺度特征, 轻量级, 目标检测

Abstract:

Bat detection in the wild is crucial for ecological protection and scientific research. To address the challenges brought by limited computing resources and complex wild environments, a lightweight bat detection model (LiteDETR-Bat) was proposed to achieve efficient real-time detection. Firstly, in order to solve the problem of feature mapping redundancy, the reparameterized convolutional efficient layer aggregation network (RCELAN) was introduced, replacing the traditional ResNet backbone network and adopting a multi-branch feature aggregation mechanism, thereby reducing computational complexity and parameter quantity. Secondly, a dynamic sampling-multi scale feature fusion (DS-MFF) was designed. This structure integrated dilated convolution and dynamic sampling operators, optimizing 0multi-scale feature fusion by expanding the receptive field and adaptively adjusting sampling positions, which enhanced the flexibility and robustness of the model in processing diversified features. Finally, a bat dataset covering various lighting conditions, perspective changes, and bat morphology changes was collected in the wild environment of Anhui Province, and related experiments such as model performance were conducted on this dataset. Experimental results showed that the proposed LiteDETR-Bat model not only reduced the number of parameters by 46.5% and achieved an mAP of 97.2%, but also made certain improvements in accuracy and real-time performance compared with the YOLO series algorithms. The LiteDETR-Bat model provided strong technical support for the monitoring and protection of wild bats, and demonstrated its application potential in ecological monitoring and biodiversity conservation.

Key words: wild bats, RT-DETR, multi-scale feature, lightweight, object detection

中图分类号:

TP391.41
Q95

王杨, 马唱, 胡明, 孙涛, 饶元, 袁振羽. 基于多尺度特征融合的轻量型野外蝙蝠检测[J]. 图学学报, 2025, 46(1): 70-80.

WANG Yang, MA Chang, HU Ming, SUN Tao, RAO Yuan, YUAN Zhenyu. Lightweight wild bat detection method based on multi-scale feature fusion[J]. Journal of Graphics, 2025, 46(1): 70-80.

图/表 17

图1 LiteDETR-Bat框架

Fig. 1 The framework of LiteDETR-Bat

图2 RCELAN结构

Fig. 2 The structure of RCELAN

图3 DS-MFF结构

Fig. 3 The structure of DS-MFF

图4 动态采样

Fig. 4 The process of dynamic sample

图5 数据集中部分蝙蝠图像示例((a)栖息在洞穴中的蝙蝠群；(b)飞行中的蝙蝠；(c)从侧面观察的蝙蝠个体；(d)从正面观察的蝙蝠个体；(e)栖息在岩壁上的蝙蝠；(f)发出声波的蝙蝠；(g)从背部观察的蝙蝠个体；(h)夜间蝙蝠群)

Fig. 5 Examples of some bat images in the dataset((a) A colony of bats roosting in a cave; (b) A bat in flight; (c) A bat observed from the side; (d) A bat observed from the front; (e) A bat roosting on a rock wall; (f) A bat emitting sound waves; (g) A bat observed from the back; (h) A group of bats at night)

表1 蝙蝠数据集信息

Table 1 The information of bat dataset

类别	数量/张
大蹄蝠(Hipposideros armiger)	614
普氏蹄蝠(Hipposideros pratti)	496
棕蝠(Myoti)	662
菊头蝠(Rousettus leschenaultia)	656
总数	2428

表2 实验参数设置

Table 2 Experimental parameter setting

实验参数	数值
训练轮次(Epoch)	100
批量大小(Batchsize)	4
线程(Workers)	4
优化器(Optimizer)	AdamW
初始学习率(lr0)	0.000 1
动量因子(Momentum)	0.9
权重衰减系数(Weight decay)	0.000 1

表3 RT-DETR与LiteDETR-Bat在蝙蝠数据集上的检测评价指标对比

Table 3 Comparison of detection evaluation indicators between RT-DETR and LiteDETR-Bat on bat dataset

方法	类别	评价指标/%
方法	类别	AP	P	R	mAP
RT-DETR^[15]	大蹄蝠 (Hipposideros armiger)	93.8	90.5	92.4	95.1
	普氏蹄蝠 (Hipposideros pratti)	94.3
	棕蝠(Myoti)	93.1
	菊头蝠 (Rousettus leschenaultia)	99.3
LiteDETR-Bat	大蹄蝠 (Hipposideros armiger)	96.4	96.0	93.0	97.2
	普氏蹄蝠 (Hipposideros pratti)	97.9
	棕蝠(Myoti)	94.9
	菊头蝠 (Rousettus leschenaultia)	99.4

图6 RT-DETR与LiteDETR-Bat的检测热力图对比

Fig. 6 Comparison of detection heatmaps between RT-DETR and LiteDETR-Bat ((a) Input image; (b) RT_DETR; (c) LiteDETR-Bat)

表4 改进模块对模型性能的影响

Table 4 Impact of the improvement module on the performance of the model

方法	RCELAN	DS-MFF	Para/M	GFLOPs	FPS/帧每秒	mAP/%
RT-DETR	-	-	18.95	57.0	99.4	95.1
RT-DETR	√	-	8.63	26.4	117.0	95.4
RT-DETR	-	√	19.26	61.5	63.4	95.5
本算法	√	√	10.12	35.6	191.3	97.2

图7 模块优化前后的参数量对比

Fig. 7 Comparison of parameters about module optimization

表5 不同模型在蝙蝠数据集上的性能对比

Table 5 Performance comparison of different models on the bat dataset

方法	Backbone	Weight/M	mAP/%	Para/M	GFLOPs	FPS/帧每秒
YOLOv5^[19]	CSPDarknet53	40.3	97.2	19.91	48.3	158.1
YOLOv6^[20]	EfficientRep	31.3	95.1	15.54	44.0	164.2
YOLOv7^[21]	CSPDarknet53	71.4	94.2	35.48	105.2	108.7
YOLOv8^[4]	CSPDarknet53	21.5	95.3	10.61	28.4	176.8
YOLOv9^[22]	-	540.1	98.4	67.03	313.4	47.9
YOLOv10^[23]	-	39.5	97.0	19.47	98.0	129.0
RT-DETR^[15]	ResNet18	38.6	95.1	18.95	57.0	99.4
RT-DETR_RCELAN	RCELAN	17.7	95.4	8.62	26.4	117.0
RT-DETR_DS-MFF	ResNet18	39.2	95.5	19.26	61.5	63.4
LiteDETR-Bat	RCELAN	20.8	97.2	10.12	35.6	191.3

图8 不同模型的mAP曲线分析

Fig. 8 Analysis of mAP curves of different models

图9 不同模型的Loss曲线分析

Fig. 9 Analysis of Loss curves of different models

图10 不同模型检测效果可视化对比

Fig. 10 Visual comparison of different model checking effects ((a) Input image; (b) YOLOv5; (c) YOLOv7; (d) YOLOv9; (e) YOLOv10; (f) RT-DETR; (g) LiteDETR-Bat)

表6 在CUB_200_2011数据集上的泛化性能对比

Table 6 Comparison of generalization performance on the CUB_200_2011 dataset

方法	P/%	R/%	mAP/%	Para/M	FPS/帧每秒
RT-DETR	71.7	65.3	66.4	19.20	78.8
本文算法	74.8	67.0	70.0	11.14	104.9

图11 CUB_200_2011数据集上的检测效果

Fig. 11 Detection results on the CUB_200_2011 dataset ((a) RT-DETR; (b) LiteDETR-Bat)

参考文献 24

[1]	FAO. Global Forest Resources Assessment 2020-key findings[R]. Rome: FAO, 2020.
[2]	胡丹, 罗正汉, 叶福强, 等. 蝙蝠携带重要病毒研究进展[J]. 中国病原生物学杂志, 2023, 18(1): 111-116.
	HU D, LUO Z H, YE F Q, et al. Advances in bat carrier of important viruses[J]. Journal of Pathogen Biology, 2023, 18(1): 111-116 (in Chinese).
[3]	PETSO T, JAMISOLA JR R S, MPOELENG D. Review on methods used for wildlife species and individual identification[J]. European Journal of Wildlife Research, 2022, 68(1): 3.
[4]	朱强军, 胡斌, 汪慧兰, 等. 基于轻量化YOLOv8s交通标志的检测[J]. 图学学报, 2024, 45(3): 422-432. DOI
	ZHU Q J, HU B, WANG H L, et al. Detection of traffic signs based on lightweight YOLOv8s[J]. Journal of Graphics, 2024, 45(3): 422-432 (in Chinese). DOI
[5]	CHEN X, ZHAO J, CHEN Y H, et al. Automatic standardized processing and identification of tropical bat calls using deep learning approaches[J]. Biological Conservation, 2020, 241: 108269.
[6]	KRIVEK G, GILLERT A, HARDER M, et al. BatNet: a deep learning‐based tool for automated bat species identification from camera trap images[J]. Remote Sensing in Ecology and Conservation, 2023, 9(6): 759-774.
[7]	XIE J J, ZHONG Y J, ZHANG J G, et al. A review of automatic recognition technology for bird vocalizations in the deep learning era[J]. Ecological Informatics, 2023, 73: 101927.
[8]	PENG J B, WANG D L, LIAO X H, et al. Wild animal survey using UAS imagery and deep learning: modified Faster R-CNN for kiang detection in Tibetan Plateau[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2020, 169: 364-376.
[9]	陈天华, 朱家煊, 印杰. 基于注意力机制的鸟类识别算法[J]. 计算机应用, 2024, 44(4): 1114-1120. DOI
	CHEN T H, ZHU J X, YIN J. Bird recognition algorithm based on attention mechanism[J]. Journal of Computer Applications, 2024, 44(4): 1114-1120 (in Chinese). DOI
[10]	苑朝, 赵亚冬, 张耀, 等. 基于YOLO轻量化的多模态行人检测算法[J]. 图学学报, 2024, 45(1): 35-46. DOI
	YUAN C, ZHAO Y D, ZHANG Y, et al. Lightweight multi-modal pedestrian detection algorithm based on YOLO[J]. Journal of Graphics, 2024, 45(1): 35-46 (in Chinese). DOI
[11]	CARION N, MASSA F, SYNNAEVE G, et al. End-to-end object detection with transformers[C]// The 16th European Conference on Computer Vision. Cham: Springer, 2020: 213-229.
[12]	HUANG Z H, LI H, XIONG X R, et al. End-to-end individual pig detection based on transfer learning[C]// The 6th International Conference on Pattern Recognition and Artificial Intelligence. New York: IEEE Press, 2023: 236-241.
[13]	LUO H T, LUO X N, LI F, et al. Identification and detection of marine mammal dorsal fin based on deformable-DETR[C]// The 13th International Conference on Information Science and Technology. New York: IEEE Press, 2023: 349-357.
[14]	李刚, 张运涛, 汪文凯, 等. 采用DETR与先验知识融合的输电线路螺栓缺陷检测方法[J]. 图学学报, 2023, 44(3): 438-447. DOI
	LI G, ZHANG Y T, WANG W K, et al. Defect detection method of transmission line bolts based on DETR and prior knowledge fusion[J]. Journal of Graphics, 2023, 44(3): 438-447 (in Chinese). DOI
[15]	ZHAO Y, LV W Y, XU S L, et al. DETRs beat YOLOs on real-time object detection[C]// 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2024: 16965-16974.
[16]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2016: 770-778.
[17]	LIU W Z, LU H, FU H T, et al. Learning to upsample by learning to sample[C]// 2023 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2023: 6027-6037.
[18]	SHI W Z, CABALLERO J, HUSZÁR F, et al. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2016: 1874-1883.
[19]	Ultralytics. YOLOv5 in PyTorch[EB/OL]. [2024-06-01]. https://github.com/ultralytics/yolov5.
[20]	LI C Y, LI L L, JIANG H L, et al. YOLOv6: a single-stage object detection framework for industrial applications[EB/OL]. (2022-09-07) [2024-05-26]. https://arxiv.org/abs/2209.02976.
[21]	WANG C Y, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]// 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2023: 7464-7475.
[22]	WANG C Y, YEH I H, LIAO H Y M. YOLOv9: learning what you want to learn using programmable gradient information[EB/OL]. (2024-02-29) [2024-05-26]. https://arxiv.org/abs/2402.13616.
[23]	WANG A, CHEN H, LIU L H, et al. YOLOv10: real-time end-to-end object detection[EB/OL]. (2024-05-23) [2024-06-01]. https://arxiv.org/abs/2405.14458.
[24]	WAH C, BRANSON S, WELINDER P, et al. The caltech-ucsd birds-200-2011 dataset[J]. California Institute of Technology, 2011, 32(1): 1-6.

基于多尺度特征融合的轻量型野外蝙蝠检测

Lightweight wild bat detection method based on multi-scale feature fusion

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 17

参考文献 24

相关文章 15

编辑推荐

Metrics

本文评价

[1]	程旭东, 史彩娟, 高炜翔, 王森, 段昌钰, 闫晓东. 面向域自适应目标检测的一致无偏教师模型[J]. 图学学报, 2025, 46(1): 114-125.
[2]	崔克彬, 耿佳昌. 基于EE-YOLOv8s的多场景火灾迹象检测算法[J]. 图学学报, 2025, 46(1): 13-27.
[3]	苑朝, 赵明雪, 张丰羿, 冯晓勇, 李冰, 陈瑞. 基于点云特征增强的复杂室内场景3D目标检测[J]. 图学学报, 2025, 46(1): 59-69.
[4]	孙前来, 林绍杭, 刘东峰, 宋晓阳, 刘佳耀, 刘瑞珍. 基于元学习的小样本指针式仪表检测方法[J]. 图学学报, 2025, 46(1): 81-93.
[5]	李琼, 考月英, 张莹, 徐沛. 面向无人机航拍图像的目标检测研究综述[J]. 图学学报, 2024, 45(6): 1145-1164.
[6]	李珍峰, 符世琛, 徐乐, 孟博, 张昕, 秦建军. 基于MBI-YOLOv8的煤矸石目标检测算法研究[J]. 图学学报, 2024, 45(6): 1301-1312.
[7]	闫建红, 冉同霄. 基于YOLOv8的轻量化无人机图像目标检测算法[J]. 图学学报, 2024, 45(6): 1328-1337.
[8]	胡凤阔, 叶兰, 谭显峰, 张钦展, 胡志新, 方清, 王磊, 满孝锋. 一种基于改进YOLOv8的轻量化路面病害检测算法[J]. 图学学报, 2024, 45(5): 892-900.
[9]	刘丽, 张起凡, 白宇昂, 黄凯烨. 结合Swin Transformer的多尺度遥感图像变化检测研究[J]. 图学学报, 2024, 45(5): 941-956.
[10]	姜晓恒, 段金忠, 卢洋, 崔丽莎, 徐明亮. 融合先验知识推理的表面缺陷检测[J]. 图学学报, 2024, 45(5): 957-967.
[11]	章东平, 魏杨悦, 何数技, 徐云超, 胡海苗, 黄文君. 特征融合与层间传递：一种基于Anchor DETR改进的目标检测方法[J]. 图学学报, 2024, 45(5): 968-978.
[12]	李建华, 韩宇, 石开铭, 张可嘉, 郭红领, 方东平, 曹佳明. 施工现场小目标工人检测方法[J]. 图学学报, 2024, 45(5): 1040-1049.
[13]	孙己龙, 刘勇, 周黎伟, 路鑫, 侯小龙, 王亚琼, 王志丰. 基于DCNv2和Transformer Decoder的隧道衬砌裂缝高效检测模型研究[J]. 图学学报, 2024, 45(5): 1050-1061.
[14]	李大湘, 吉展, 刘颖, 唐垚. 改进YOLOv7遥感图像目标检测算法[J]. 图学学报, 2024, 45(4): 650-658.
[15]	张新宇, 张家意, 高欣. ASC-Net：腹腔镜视频中手术器械与脏器快速分割网络[J]. 图学学报, 2024, 45(4): 659-669.