Mushroom is highly diverse in morphology and colors, and difficult for ordinary people to discriminate between them. The lack of high-quality labeled mushroom datasets is one of the bottlenecks restricting cutting-edge image recognition models to achieve high performance in mushroom image recognition. To address the limitation, in this paper, we will introduce a large mushroom dataset, called Mushroom-23, constructed by us. It collects over 35000 labeled mushroom images belonging to 203 popular mushroom species in Hong Kong. Along with constructing the new mushroom dataset, we also propose the domain adaptive pre-trained (DAPT) model to make the state-of-the-art Vision Transformer (ViT) adaptive to specific mushroom recognition. The DAPT is first pre-trained on ImageNet and Danish Fungi (DF20) datasets, then fine-tuned on the collected Mushroom-23 dataset to gain the capability of different categories of mushrooms. Extensive experimental results show DAPT outperforms all baseline models by a large margin in terms of Accuracy and Macro F1.
|Title of host publication||Advanced Data Mining and Applications :19th International Conference, ADMA 2023, Shenyang, China, August 21–23, 2023, Proceedings, Part IV|
|Editors||Xiaochun YANG, Heru SUHARTANTO, Guoren WANG, Bin WANG, Jing JIANG, Bing LI, Huaijie ZHU, Ningning CUI|
|Number of pages||13|
|Publication status||Published - 2023|
|Event||19th International Conference on Advanced Data Mining and Applications, ADMA 2023 - Shenyang, China|
Duration: 21 Aug 2023 → 23 Aug 2023
|Name||Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)|
|Conference||19th International Conference on Advanced Data Mining and Applications, ADMA 2023|
|Period||21/08/23 → 23/08/23|
Bibliographical notePublisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.
- Mushroom Recognition
- Domain Adaption
- Pre-trained Model
- Vision Transformer