Automatically Detecting Zokor Molehills in Forests Using Machine Learning

Keqin Song; Zhiwei Wang; Limin Wang; Jihui Hao; Lijie Wang; Tianliang Zhang

doi:10.23977/jipta.2026.090103

Automatically Detecting Zokor Molehills in Forests Using Machine Learning

Download as PDF

DOI: 10.23977/jipta.2026.090103 | Downloads: 2 | Views: 57

Author(s)

Keqin Song ¹, Zhiwei Wang ¹, Limin Wang ¹, Jihui Hao ¹, Lijie Wang ¹, Tianliang Zhang ^2,3

Affiliation(s)

¹ Weichang Manchu and Mongol Autonomous County State-Owned Luanhe Forest Farm, Weichang County, Chengde, China
² Institute of Applied Mathematics, Hebei Academy of Sciences, No. 46 South Youyi Street, Shijiazhuang, China
³ Hebei Information Security Certification Technology Innovation Center, No. 46 South Youyi Street, Shijiazhuang, China

Corresponding Author

Tianliang Zhang

ABSTRACT

Traditional monitoring of zokor mounds relies primarily on manual ground surveys, which suffer from low efficiency, high costs, significant limitations due to terrain and vegetation cover, high rates of missed observations, and strong subjectivity in data collection. These methods struggle to meet the demands of modern forestry for precise, large-scale, and dynamic monitoring. This study focuses on the Saihanba Forest Area in Hebei Province, China, and develops an automated detection system for zokor mounds based on UAV visible-light imagery and the YOLOX deep learning model. A DJI Mavic 3 drone was used to acquire visible-light remote sensing imagery at a flight altitude of 30 m. A dedicated dataset for zokor mounds in the forest area was constructed, comprising 862 valid images and 1,286 annotated samples, covering targets of varying habitats and sizes. YOLOX was selected as the core detection model, and transfer learning based on COCO pre-trained weights was employed for model training and parameter optimization to enhance convergence speed and generalization ability. Experimental results show that the optimized YOLOX model achieved an average precision (AP) of 99.90% on the test set, enabling accurate identification of zokor mounds against complex forest backgrounds with excellent resistance to false positives and false negatives; the model’s precision-recall (PR) curve lies predominantly in the upper-right quadrant of the graph, maintaining stable and efficient detection performance across different confidence thresholds. This study demonstrates the feasibility and advantages of using visible-light imagery from consumer-grade drones combined with deep learning methods for the rapid, large-scale monitoring of zokor mounds in forested areas. This approach provides efficient, low-cost technical support for the precise prevention and control of rodent damage, the dynamic assessment of forest ecosystems, and the scientific evaluation of the ecological impacts of zokors, and holds significant theoretical value and promising prospects for engineering applications.

KEYWORDS

UAV remote sensing; YOLOX; Object detection; Zokor mounds; Forest monitoring

CITE THIS PAPER

Keqin Song, Zhiwei Wang, Limin Wang, Jihui Hao, Lijie Wang, Tianliang Zhang. Automatically Detecting Zokor Molehills in Forests Using Machine Learning. Journal of Image Processing Theory and Applications (2026) Vol. 9, No.1, 22-30. DOI: http://dx.doi.org/10.23977/jipta.2026.090103.

REFERENCES

[1] Li S C.(2025). Assessment of Forest Tree Damage Caused by Myospalax baileyi and Screening of Ｒodenticides in the Changling Mountain Forest Area. Forest Investigation Design,54(05),28-32.
[2] Song X N, Li L, Lu S H, et al.(2025). The Impact of Rodents in Heilongjiang's Forest Areas on Cash Crops and Forest Regeneration. Rural Economy and Science-Technology,36(15),81-84.
[3] Liu J X, Xie S Q, Zhang Z H, et al.(2020). Study on spatial sidtribution and sampling technique of major rodent pest in Saihanba area. Foretry and Ecological Sciences,35(02),186-190.
[4] Niu, Y, Yang, S, Zhu, H et al.(2020). Cyclic formation of zokor mounds promotes plant diversity and renews plant communities in alpine meadows on the Tibetan Plateau. Plant Soil,446, 65–79.
[5] Xiang, Z, Bhatt, A, Tang, Z, et al. (2021) Disturbance of plateau zokor-made mound stimulates plant community regeneration in the Qinghai-Tibetan Plateau, China. J. Arid Land 13, 1054–1070.
[6] Su, F, Wang, F, Li, D, et al(2025). Plateau zokor mounds decrease soil respiration by shifting microenvironment. CATENA 258.
[7] Ge, Z, Liu, S, Wang, F, et al(2021). Yolox: exceeding yolo series in 2021. ArXiv, abs/2107.08430.
[8] Han Y, Liu H R, Lin W S.(2025). Tree Species Identification Based on Improved YOLOv10 and UAV RGB Imagery. FOREST ENGINEERING,41(05),922-935.

Subscription

E-Mail Alert

Downloads:	3025
Visits:	236673

Automatically Detecting Zokor Molehills in Forests Using Machine Learning

Author(s)

Affiliation(s)

Corresponding Author

ABSTRACT

KEYWORDS

CITE THIS PAPER

REFERENCES

RESOURCES

JOIN US

PUBLICATION SERVICES

CONTACT US