Classification of Blueberry Varieties Based on Improved EfficientNet
2022 4th International Academic Exchange Conference on Science and Technology Innovation (IAECST)(2022)
摘要
A method to classify blueberry varieties based on an improved EfficientNet is proposed, aiming at the high cost of manual classification and the difficulty of manual confirmation of new varieties. First, considering that there are few datasets of blueberry leaf samples available on the internet, blueberry leaf samples from different seasons are collected and photographed, and four blueberry leaf image datasets are established, including germination, maturity, dormancy, and mixed stages. Second, to better allocate feature weights, an improved EfficientNet is introduced to generate a blueberry variety classification model. Finally, experiments on different datasets are conducted using different models for blueberry variety classification. For the recognition results of the dormant-stage dataset, the accuracy rate of the improved EfficientNet method reaches 99.85%, exceeding EfficientNet-B1 by 3.05%. For the recognition results of datasets of the germination, maturity, dormancy, and mixed stages, the accuracy rates of the improved EfficientNet method are 98.50%, 99.65%, 99.85%, and 97.92%, respectively. These results show that the improved EfficientNet can be applied to blueberry variety recognition with a high recognition rate, and it has a certain robustness to different periods of leaves. This research can provide new inspiration and technical support for blueberry variety recognition.
更多查看译文
关键词
component,blueberry variety identification,leaf images,EfficientNet-B1,CBAM
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络