针对传统人工特征提取模型难以满足复杂场景下目标识别的需求,提出了一种基于CLMF 的深度卷积神经网络(Convolutional neural networks with candidate location and multi-feature fusion, CLMF-CNN)。该模型结合视觉显著性、多特征融合和CNN模型实现目标对象的识别。首先,利用加权Itti模型获取目标候选区；然后,利用CNN模型从颜色、亮度多特征角度提取目标对象的特征,经过加权融合供目标识别；最后,与单一特征以及目前的流行算法进行对比实验,结果表明本文模型不仅在同等条件下正确识别率得到了提高,同时,达到实时性要求。%To solve the problem that the traditional manual feature extraction models are unable to satisfy object recognition in complex environment, an object recognition model based on convolutional neural networks with candidate location and multi-feature fusion (CLMF-CNN) model is proposed. The model combines the visual saliency, multi-feature fusion and CNN model to realize the object recognition. Firstly, the candidate objects are conformed via weighted Itti model. Consequently, color and intensity features are obtained via CNN model respectively. After the multi-feature fusion method, the features can be used for object recognition. Finally, the model is tested and compared with the single feature method and current popular algorithms. Experimental result in this paper proves that our method can not only get good performance in improving the accuracy of object recognition, but also satisfy real-time requirements.
展开▼
