【Objective】As a core element of traditional Chinese gardens, Taihu rockeries are long exposed to natural environments and susceptible to deterioration caused by weathering, biological attachment, and structural stress, which threatens their heritage value. To address the low efficiency and high subjectivity of traditional manual visual inspection, this study aims to develop a high-precision and high-efficiency intelligent recognition model for surface deterioration of Taihu rockeries.【Method】Taking Taihu rockeries from typical gardens in Suzhou and Yangzhou as research objects, an improved YOLOv12-based deterioration detection method was proposed. Images of Taihu rockeries were collected to construct a dataset covering four deterioration types: cement repair, stress cracking, biological attachment, and crusting. The BiFPN (bidirectional feature pyramid network) multi-scale feature fusion module was introduced into the YOLOv12 architecture to replace the original PAN structure, thereby enhancing the model's ability to represent multi-scale deterioration targets under complex texture backgrounds. The performance of the proposed model was validated through ablation experiments, comparative experiments, and field application tests.【Result】Experimental results show that the improved model achieved a mAP@0.5 of 0.856 and a mAP@0.5-0.95 of 0.662 on the test set, representing increases of 4.0% and 4.2%, respectively, compared with the baseline model, and outperforming mainstream models such as YOLOv8 and YOLOv11. In the field application at the Nine Lions Rockery of Zhou's Xiaopangu in Yangzhou, the model achieved an overall recognition accuracy of 83.82% and a detection speed of approximately 30 frames per second (33.6 ms per image), with recognition accuracies reaching 100% for crusting and 90% for biological attachment.【Conclusion】The introduction of the BiFPN module effectively enhances the model's adaptability to the complex scenes of Taihu rockeries, enabling rapid and accurate deterioration detection. The model combines high precision and real-time performance, serving as a reliable auxiliary tool for professionals and providing robust technical support for the digital conservation and preventive monitoring of garden heritage.
Key words
garden rockery /
deterioration detection /
YOLOv12 /
BiFPN /
deep learning
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
References
[1] Thacker C,Keswick M.The Chinese garden:history,art and architecture[J].Garden History,1979,7(1):20. DOI:10.2307/1586713.
[2] Dong Q L,Wei T H,Zhang Q P,et al.The texture of Chinese garden rockery stones:based on 3D point cloud and 3D printing technology[J].npj Heritage Science,2025,13:47. DOI:10.1038/s40494-025-01582-5.
[3] 张青萍,王岑岑,张晓东.基于有限元分析的园林假山遗产病害监测与结构性保护研究[J].南京林业大学学报(自然科学版),2024,48(6):252-259.Zhang Q P,Wang C C,Zhang X D.Research on the disease monitoring and structural protection of garden rockery heritage based on an finite element analysis[J].Journal of Nanjing Forestry University (Natural Sciences Edition),2024,48(6):252-259. DOI:10.12302/j.issn.1000-2006.202207035.
[4] Fu L,Zhi H,Zhang Q P.A risk index for the structural diagnosis of rockeries in Chinese classical gardens based on the MIVES model[J].npj Heritage Science,2025,13:37. DOI:10.1038/s40494-025-01613-1.
[5] Chen W Y,Qin K.Rockery protection under the framework of world heritage protection[J].Journal of Landscape Research,2019,11(1):61-64. DOI: 10.16785/j.issn1943-989x.2019.1.013
[6] Mansuri L E,Patel D A.Artificial intelligence for heritage conservation:a case study of automatic visual inspection system[M]//Current State of Art in Artificial Intelligence and Ubiquitous Cities.Singapore:Springer,2022:1-15. DOI:10.1007/978-981-19-0737-1_1.
[7] Mishra M,Lourenço P B.Artificial intelligence-assisted visual inspection for cultural heritage:state-of-the-art review[J].Journal of Cultural Heritage,2024,66:536-550. DOI:10.1016/j.culher.2024.01.005.
[8] Zhang L,Chen Y L,Zheng L,et al.Investigating the surface damage to Fuzhou’s ancient houses (gu-Cuo) using a non-destructive testing method constructed via machine learning[J].Coatings,2024,14(11):1466. DOI:10.3390/coatings14111466.
[9] Idjaton K,Janvier R,Balawi M,et al.Detection of limestone spalling in 3D survey images using deep learning[J].Automation in Construction,2023,152:104919. DOI:10.1016/j.autcon.2023.104919.
[11] Yan L N,Chen Y L,Zheng L,et al.Application of computer vision technology in surface damage detection and analysis of shedthin tiles in China:a case study of the classical gardens of Suzhou[J].Heritage Science,2024,12:72. DOI:10.1186/s40494-024-01185-6.
[12] Karimi N,Mishra M,Lourenço P B.Deep learning-based automated tile defect detection system for Portuguese cultural heritage buildings[J].Journal of Cultural Heritage,2024,68:86-98. DOI:10.1016/j.culher.2024.05.009.
[13] Zheng L,Chen Y L,Yan L N,et al.Automatic detection and recognition method of Chinese clay tiles based on YOLOv4:a case study in Macau[J].International Journal of Architectural Heritage,2024,18(10):1551-1570. DOI:10.1080/15583058.2023.2246029.
[14] Terven J,Córdova-Esparza D M,Romero-González J A.A comprehensive review of YOLO architectures in computer vision:from YOLOv1 to YOLOv8 and YOLO-NAS[J].Machine Learning and Knowledge Extraction,2023,5(4):1680-1716. DOI:10.3390/make5040083.
[15] Tian Y J,Ye Q X,Doermann D.YOLOv12:attention-centric real-time object detectors[PP/OL].arXiv (2025-02-18)[2025-11-16].DOI: 10.48550/arXiv.2502.12524.
[16] Du F J,Jiao S J.Improvement of lightweight convolutional neural network model based on YOLO algorithm and its research in pavement defect detection[J].Sensors,2022,22(9):3537. DOI:10.3390/s22093537.
[17] 傅力,董芊里,张青萍.江南园林假山遗产病害分类研究[J].建筑遗产,2023(4):52-58.Fu L,Dong Q L,Zhang Q P.Pathological categorisation of heritage rockeries in Jiangnan gardens[J].Heritage Architecture,2023(4):52-58. DOI:10.19673/j.cnki.ha.2023.04.007.
[18] Dong Q L,Wei T H,Wang Y,et al.Intangible cultural heritage based on finite element analysis:force analysis of Chinese traditional garden rockery construction[J].Heritage Science,2024,12:241. DOI:10.1186/s40494-024-01355-6.
[19] 张青萍,陈凤仪.植物根劈作用对江南园林假山遗产的影响:以南京瞻园为例[J].南京林业大学学报(自然科学版),2024,48(4):271-278.Zhang Q P,Chen F Y.The influence of plant root splitting on garden rockeries in Jiangnan:taking the Zhanyuan Garden as an example[J].Journal of Nanjing Forestry University (Natural Sciences Edition),2024,48(4):271-278. DOI:10.12302/j.issn.1000-2006.202301025.
[20] Cappitelli F,Toniolo L,Sansonetti A,et al.Advantages of using microbial technology over traditional chemical technology in removal of black crusts from stone surfaces of historical monuments[J].Applied and Environmental Microbiology,2007,73(17):5671-5675. DOI:10.1128/AEM.00394-07.
[21] Khanam R,Hussain M.YOLOv11:an overview of the key architectural enhancements[J].ArXiv Preprint ArXiv (2024-10-23)[2025-11-16]. DOI: 10.48550/arXiv.2410.17725
[22] Wang C Y,Bochkovskiy A,Liao H 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 (CVPR).June 17-24,2023.Vancouver,BC,Canada.IEEE,2023:7464-7475. DOI:10.1109/cvpr52729.2023.00721.
[23] Tan M X,Pang R M,Le Q V.EfficientDet:scalable and efficient object detection[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).June 13-19,2020,Seattle,WA,USA.IEEE,2020:10778-10787. DOI:10.1109/CVPR42600.2020.01079.
[24] Liu S,Qi L,Qin H F,et al.Path aggregation network for instance segmentation[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.June 18-23,2018.Salt Lake City,UT.IEEE,2018:8759-8768. DOI:10.1109/cvpr.2018.00913.
[25] 张青萍,巫文笑.苏州园林与扬州园林的差异:以小盘谷为例[J].园林,2018,35(3):8-11.Zhang Q P,Wu W X.Differences between Suzhou gardens and Yangzhou gardens:taking xiaopan valley as an example[J].Landscape Architecture,2018,35(3):8-11.
PDF(907 KB)
