集成学习与迁移学习的作物病害图像识别算法

Crop disease image recognition algorithm based on ensemble learning and transfer learning

为提高包含复杂背景信息的作物病害图像的识别准确率,解决作物病害数据集样本较少而出现的模型训练过拟合问题,提出了一种基于集成学习与迁移学习方法的作物病害图像识别算法。该算法首先在公开数据集上完成模型预训练,其次通过任务域迁移和特征空间迁移,解决农作物病害图像识别问题;进而重构集成学习中的投票机制算法,提升模型对复杂图像的识别能力。实验结果表明,该算法识别准确率为98.324%,可较好地实现对复杂背景信息图像的准确识别。

In order to improve the recognition accuracy of crop disease image with complex background information and solve the problem of over-fitting in the process of model training when there are few samples in crop disease dataset,a crop disease image recognition algorithm based on ensemble learning and transfer learning was proposed.After the model pre-training was completed on the open datasets,the task domain and feature space of the model were transferred to the dataset of crop diseases by transfer learning,and the recognition ability of the model for complex images was improved by reconstructing the voting mechanism in ensemble learning.The experimental results show that the recognition accuracy of the proposed algorithm is 98.324%,which can realize the accurate recognition of the complex background information images.