Authors

Abstract

In view of the difficulty and high cost of monitoring the invasion of small aggregations of Spartina alterniflora in coastal wetlands, this study proposes a SA-YOLO detection model. First, by adopting a lightweight cascade attention mechanism as the feature extraction part of the network, the model's ability to extract features from Spartina alterniflora images is optimized. Secondly, the convolution layer with an improved adaptive attention mechanism is added to optimize feature extraction, dynamically adjust the weight of the feature map, and reduce the amount of calculation. Thirdly, the improved adaptive convolution network is used to optimize the original neck layer, improve the model's ability to integrate Spartina alterniflora image features, and reduce the amount of calculation. Finally, a Spartina alterniflora recognition system is independently built. The system effectively implements the proposed method and realizes the detection and recording of Spartina alterniflora information. This study successfully verifies the effectiveness of the proposed method by conducting experiments on the actual collected Spartina alterniflora dataset. The test results show that the recall rate and accuracy of the proposed SA-YOLO Spartina alterniflora detection model are 94.5% and 92.4%, respectively, both reaching a high level. It can be seen that the model can complete the identification and detection tasks of Spartina alterniflora, providing a solution for the identification and information collection of Spartina alterniflora in coastal areas.

Abstract in Chinese

为实现对复杂环境中互花米草的快速识别和信息采集，及时发现和清理，保护生态环境的目的。针对互花米草的生长环境和植物特性，本研究提出了一种基于级联注意力机制和可变形卷积的SA-YOLO互花米草识别方法。首先，通过采用级联注意力机制作为网络的特征提取部分，优化模型对于互花米草图像的识别精度和特征提取的能力，提高模型识别速度和精度；其次，添加自适应注意力机制改进型的卷积层，优化特征提取，动态的调整特征图的权重和减少计算量；最后，采用改进的自适应卷积网络优化原neck层，提升模型对互花米草图像特征的融合能力并减少计算量。测试结果表明：改进型SA-YOLO的互花米草检测模型的召回率和精确度分别为94.5%和92.4%，均达到较高水平。由此可知，该模型可以完成对互花米草的识别检测任务，并表现出良好的鲁棒性和实时性，为滨海地区互花米草的识别和信息采集提供了解决方案。