Comparison of YOLO instance segmentation models in identifying forest species using leaf images
Resumo
The vast area of forests in Brazil, along with the biodiversity of the flora, poses a challenge to the identification of tree species. This process can be time-sensitive due to the nature of certain projects; therefore, the use of leaves for the identification is ideal in such cases. In order to assist this process, automated tools that use deep learning can be an aid in the conservation efforts. This study evaluates and compares two YOLO (You Only Look Once) models, the nano versions of YOLOv11 and YOLOv12, for the task of identifying species using their leaves. For this, a custom dataset was created with 26 species under varied conditions; the images also have a complex background to guarantee the model’s efficiency in fieldwork. Overall, YOLOv11 achieved higher recall (56.746%) and mAP@50 (72.634%), YOLOv12 demonstrated superior precision (86.054%) and better per-class performance. The results show that both models’ performance varies due to the dataset annotation and the leaf morphology, as both models struggled with species exhibiting compound leaves. This work contributes to the development of practical tools for biodiversity monitoring and supports future advancements in automated species identification, highlighting the potential of lightweight models for tree species identification.
Referências
——, “The second report on the state of the world’s forest genetic resources,” Rome, 2025. [Online]. DOI: 10.4060/cd4838en
R. C. V. M. d. Silva, M. J. G. Hopkins, and I. S. Thompson, “Identificação botânica na amazônia: Situação atual e perspectivas,” p. 81, 2003. [Online]. Available: [link]
N. Haider, N. Haider (2011). Identification of Plant Species using Traditional and Molecular-based Methods, pp. 1-62. In: Wild Plants: Identification, Uses and Conservation (Ed. RE Davis). Nova Science Publishers, Inc. Nova Science Publishers, 01 2011.
C. Urbanetz, J. Y. Tamashiro, and L. S. Kinoshita, “Chave de identificação de espécies lenhosas de um trecho de floresta ombrófila densa atlântica, no sudeste do brasil, baseada em caracteres vegetativos,” Biota Neotropica, vol. 10, no. 2, p. 349–398, 06 2010. [Online]. DOI: 10.1590/s1676-06032010000200036
X. Yang, A. Chen, G. Zhou, J. Wang, W. Chen, Y. Gao, and R. Jiang, “Instance segmentation and classification method for plant leaf images based on isc-mrcnn and aps-dccnn,” IEEE Access, vol. 8, pp. 151 555–151 573, 2020. [Online]. DOI: 10.1109/ACCESS.2020.3017560
R. Guo, L. Qu, D. Niu, Z. Li, and J. Yue, “Leafmask: Towards greater accuracy on leaf segmentation,” in 2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), 2021, pp. 1249–1258. [Online]. DOI: 10.1109/ICCVW54120.2021.00145
T. Yang, S. Zhou, Z. Huang, A. Xu, J. Ye, and J. Yin, “Urban street tree dataset for image classification and instance segmentation,” Computers and Electronics in Agriculture, vol. 209, p. 107852, 2023. [Online]. Available: [link]
P. Kumar, S. V. Kumar, and S. S. Pandi, “Cnn and edge-based segmentation for the identification of medicinal plants,” in 2024 5th International Conference on Intelligent Communication Technologies and Virtual Mobile Networks (ICICV), 2024, pp. 89–94. [Online]. DOI: 10.1109/ICICV62344.2024.00021
Z. Wang, X. Su, and S. Mao, “Single-species leaf detection against complex backgrounds with yolov5s,” Forests, vol. 15, 6 2024. [Online]. Available: [link]
L. D. B. C. Sabbi, C. D. Câmara, R. Cielo-Filho, A. L. d. Silva, and M. D. Thrun, “Áreas verdes urbanas: uma proposta para conservação da biodiversidade e educação ambiental, estudo de caso do bosque da utfpr, câmpus medianeira,” Tempo da Ciência, vol. 31, no. 61, p. 16, nov. 2024. [Online]. Available: [link]
J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You only look once: Unified, real-time object detection,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 779–788. [Online]. DOI: 10.1109/CVPR.2016.91
G. Jocher and J. Qiu, “Ultralytics yolo11,” 2024. [Online]. Available: [link]
Y. Tian, Q. Ye, and D. Doermann, “Yolov12: Attention-centric real-time object detectors,” arXiv preprint arXiv:2502.12524, 2025.
