基于改进的SAE和DCT的自适应无人机巡线图像识别算法研究

Research on adaptive UAV patrol image recognition algorithm based on improved SAE and DCT

针对无人机巡线过程中无人机拍摄的航拍图像具有场景复杂多变、纹理信息丰富、容易出现图像识别结果错误的问题,本文提出了一种基于改进的稀疏自动编码器和离散余弦变换的自适应无人机巡线图像识别算法。首先,结合SAE和DCT优点构建了DCT-DSAE算法模型,然后根据DCT能量集中的特点提出了自适应DCT系数选择法,在不损失原始输入信息的前提下降低输入数据的冗余信息,最后将自适应选择的DCT系数输入到DCT-SAE模型中,通过特征学习得到图像深层次的特征表达以及最后的分类结果。实验表明,本文提出的基于改进的稀疏自动编码器和离散余弦变换的自适应无人机巡线图像识别算法识别平均正确率达到92.69%,DCT-DSAE算法能自动地学习图像深层复杂、抽象特征,重组的特征有效的提高了无人机巡线图像的识别准确率。

In order to solve the problem that the aerial images captured by the drone during the UAV line-traveling process are complex and changeable,rich in texture information,and prone to wrong image recognition results,this paper proposes an improved sparse automatic encoder and discrete cosine transform Adaptive image recognition algorithm for unmanned aerial vehicles.First,combined with the advantages of SAE and DCT,a DCT-DSAE algorithm model was constructed,and then an adaptive DCT coefficient selection method was proposed according to the characteristics of DCT energy concentration,and the redundant information of the input data was reduced without losing the original input information.Finally,the adaptively selected DCT coefficients are input into the DCT-SAE model,and deep-level feature expression and final classification results of the image are obtained through feature learning.Experiments show that the adaptive drones line recognition image recognition algorithm based on improved sparse automatic encoder and discrete cosine transform proposed in this paper has an average recognition rate of 92.69%.The DCT-DSAE algorithm can automatically learn deep complex and abstract features of the image The reorganized features effectively improve the recognition accuracy of the UAV line patrol image.