基于SceneCaps的场景图像识别

Scene Image Recognition Based on SceneCaps

场景识别是计算机视觉领域重要的研究方向。采用传统神经网络的场景识别方法对数据需求量较大、训练时间较长且不能捕捉图像的空间关系,对复杂的场景图像识别有一定的局限性。针对这一问题,在CapsNet(胶囊网络)的基础上,提出SceneCaps模型,并在Scene15和Places2数据集上进行实验。实验表明,与CapsNet模型相比,SceneCaps模型在Scene15和Places2数据集上的识别精度分别平均提高了0.25%和0.03%,训练速度分别平均提高了29%和69%,验证了该模型的有效性。

Scene recognition is an important research direction in the field of computer vision.The traditional scene recognition method based on neural network requires a lot of data,takes a long time to train and cannot capture the spatial relationship of images.To address this problem,a SceneCaps model based on capsule network(CapsNet)was proposed in this paper,and experiments were conducted on Scene15 and Places2 data sets.The experimental results show that compared with CapsNet model,the recognition accuracy rates of SceneCaps model on Scene15 and Places2 datasets are increased by 0.25%and 0.03%respectively,and the training speeds are increased by 29%and 69%respectively.