融合多尺度和迁移学习的蝴蝶种类识别

Butterfly Species Recognition Integrated with Multiscale and Transfer Learning

摘要: 蝴蝶在生态系统稳定中发挥着重要作用，不仅能帮助植物传播花粉，还能对其生存环境变化做出指示。针对自然环境中蝴蝶种类识别率低的问题，本文提出一种融合多尺度和迁移学习的识别模型。首先，使用焦点损失函数解决数据集分布不平衡问题；其次，引入迁移学习提升识别准确率、加快模型收敛；最后，引入空洞空间卷积池化金字塔，提取蝴蝶图像不同尺度信息。实验结果表明，本研究所提方法平均识别准确率达到98.03%，较原始模型提升了4.88%，相较其他对比模型也取得较大优势，可为自然环境中蝴蝶种类识别提供技术支持。

Abstract: Butterflies play an important role in ecosystem stability, not only helping plants spread pollen, but also signaling changes to their environment. Aiming at the low recognition rate of butterfly species in the natural environment, this paper proposes a recognition model that integrates multi-scale and transfer learning. Firstly, the focal loss function is used to solve the problem of imbalanced dataset distribution. Secondly, transfer learning is introduced to improve the recognition accuracy and speed up the model convergence. Finally, ASPP is introduced to enable the model to capture butterfly image information from different scales. The experimental results show that the average recognition accuracy of the method proposed in this study reaches 98.03%, which is 4.88% higher than the original model. Compared with other comparative models, it also achieves great advantages. It can provide technical support for the identification of butterfly species in the natural environment.