TY  - JOUR
AU  - ZHAO, Jiafan
AU  - LI, Xin
AU  - LI, Lijun
AU  - ZHONG, Chao
AU  - WANG, Shaoli
PY  - 2025
TI  - Recognition of Camellia oleifera anthracnose disease based on preference information and improved YOLOv8n
JO  - Journal of Central South University of Forestry & Technology
SN  - 1673-923X
SP  - 164
EP  - 176
VL  - 45
IS  - 11
AB  - 【Objective】As one of the major diseases affecting the yield and quality of oil tea, the efficient and accurate detection of oil tea anthracnose is crucial to ensure the healthy operation of oil tea bases. In order to improve the detection rate of oil tea anthracnose targets in oil tea bases, and to enhance the daily maintenance of oil tea and thus increase productivity, this paper adopted an improved preference YOLOv8n algorithm to detect oil tea anthracnose.【Method】Firstly, the color features of the infected part of anthracnose fruits are used as pre-processing information for preference information, which increases the richness of the samples while enhancing the generalization performance of the model; Secondly, the lightweight ShuffleNetv2 network was used instead of the original backbone network to optimize the performance of the network, which effectively reduced the complexity of the model and improves the detection efficiency; Thirdly, the attention mechanism, ECA-Net, is introduced to enhance the quality and expression ability of the overall features to improve the accuracy of the model; Fourth, use BiFPN structure in the neck network instead of the original network structure to improve the fusion rate of the model to the target features; Finally, DIoU is comprehensively selected in the original algorithm of YOLOv8n to strengthen the recognition ability of the occluded targets and improve the recognition effect of the oil tea anthracnose targets under the complex scenarios of occlusion and overlapping.【Result】After experimental verification, the improved YOLOv8n algorithm performs well in the oil tea anthrax detection task. Based on the input preference images, the improved algorithm achieved 87.6% image recognition precision, 80.7% recall, 81.5% F1 score, and 78.8% mean average precision; the precision increased by 3.9%, the recall increased by 2.4%, the F1 score increased by 81.5%, and the mean average precision increased by 1.4%. These indicators are better than traditional methods such as Faster R-CNN, YOLOv3, YOLOv7, YOLOXs and the original YOLOv8n model, the improved algorithm synthesized in the overall detection time and precision in line with the requirements of the oil tea disease detection, for the deployment of intelligent disease detection in agroforestry and its application provides a useful reference.【Conclusion】The improved YOLOv8n algorithm proposed in this paper has achieved remarkable results in the detection of anthracnose in oil tea, lightweight model at the same time in the detection time to achieve high efficiency, but also in the detection of accuracy to a high level. The successful application of this algorithm not only provides a powerful tool for the intelligent detection of oil tea diseases, but also provides a useful reference for the automated monitoring, prevention and control of other diseases in the field of agriculture and forestry.
UR  - https://doi.org/10.14067/j.cnki.1673-923x.2025.11.016
DO  - 10.14067/j.cnki.1673-923x.2025.11.016