文献类型： 外文期刊

作者： Peng, Ying shu ¹ ; Wang, Yi ⁴ ;

作者机构： 1.Hunan Acad Agr Sci, Hort Res Inst, Changsha 410125, Peoples R China

2.Yuelu Mt Lab, Changsha, Peoples R China

3.Chinese Acad Sci, Lushan Bot Garden, Beijing, Jiangxi, Peoples R China

4.Jiangsu Adv Energy Storage Technol Co Ltd, Battery Technol Dept, Nanjing, Peoples R China

关键词： Self-supervised learning; masked image modeling; agricultural classification tasks; vision transformer; histogram of oriented gradients feature

期刊名称：COGENT FOOD & AGRICULTURE （ 影响因子：2.3； 五年影响因子：2.9 ）

ISSN： 2331-1932

年卷期： 2025 年 11 卷 1 期

页码：

收录情况： SCI

摘要： Image classification poses a significant challenge in agriculture. However, the utilization of popular algorithms such as vision transformers and convolutional neural networks has often fallen short in numerous agricultural classification tasks owing to the scarcity of extensively labelled data and the reliance on pretrained models trained on generic datasets. To address this, our study details the pretraining of ViTs using 224,228 agricultural images, employing masked image modeling for preprocessing. The pretrained model was then fine-tuned on three independent agricultural classification datasets and performed better than state-of-the-art methods. For example, our method achieved the highest accuracy rates of 76.18%, 98.49%, and 88.56% on IP102, DeepWeeds, and Tsinghua Dogs datasets, respectively. This enhancement in accuracy can be attributed to the robust modeling strategy we have developed through extensive experimentation with the pretrained MIM model. Our modeling strategy encompasses employing advanced MIM models, leveraging the histogram of oriented gradient features as the reconstruction target, and selecting an appropriate mask ratio. We hope that this research will prompt the application of self-supervised learning techniques, represented by the MIM model, to a wide range of agricultural image-related tasks in the future.