基于深度学习和GB-RBM的UAV红外语义分割方法

Semantic segmentation of UAV infrared images based on deep learning and GB-RBM

为提高UAV红外图像语义分割的性能,提出基于深度学习和高斯伯努利受限玻尔兹曼机(GB-RBM)的实时语义分割模型。确认地面车辆实时特征提取中的关键问题。基于GB-RBM,提出用于编码阶段的形状先验模型。通过将SegNet中的编码器-解码器结构与GB-RBM模块相融合,在解码器块中生成红外数据的实时映射,实现准确快速的语义分割。实验结果表明,所提方法能够很好地处理红外视频中的实时几何信息,在3个实验数据集上的平均精度约为0.98,平均处理时长约为17.86 s,性能优于其它优秀方法。

To improve the performance of UAV infrared image semantic segmentation,a real-time semantic segmentation model based on deep learning and Gaussian-Bernoulli restricted Boltzmann machine(GB-RBM)was proposed.The key issues related to real-time feature extraction of ground vehicles were identified.A GB-RBM model based shape prior model for encoding block was proposed.The real-time mapping of the thermal data in the decoder block was realized through the fusion of encoder-decoder structure of the Segnet architecture and the GB-RBM module,eventually leading to accurate and fast semantic segmentation.Experimental results show that the proposed method can properly process the real-time geometric information in the infrared vi-deos.The mean precision and the average processing time on all datasets when using the proposed method are 0.98 and 17.86 s,respectively,which outperform that using other state-of-art methods.