摘要
水下自动目标识别一直是具有挑战性的任务.针对海洋环境下目标图像数据获取困难,样本数量不足以训练深层神经网络这一问题,提出小样本情况下基于深度学习的水下图像识别方法.利用提出的改进中值滤波器抑制水下小样本集图像的脉冲噪声;然后,采用ImageNet图像数据集对搭建的深度卷积神经网络模型进行预训练;使用水下降噪图像对经过预训练的神经网络进行参数微调.利用海洋鱼类图像数据集对完成训练的卷积网络性能进行验证,取得85.08%的正确识别率.
Automatic underwater target recognition has always been a challenging task. Aiming at the problem that it is difficult to acquire target image data in marine environment and the number of samples is not enough to train deep neural networks, an underwater image recognition method based on depth learning in the case of small samples is proposed. Firstly, the proposed improved median filter was used to suppress the impulse noise of underwater small sample set images. Then, ImageNet image data set was used to pre-train the built deep convolution neural network model. Finally, the pre-trained neural network was fine-tuned with underwater noise reduction images.. The performance of the trained convolution network was verified by using the marine fish image data set, and the correct recognition rate was 85.08 %.
引文
[1] 乔曦.基于水下机器视觉的海参实时识别研究[D].北京:中国农业大学,2017.
[2] 汤春瑞.水下目标声图像多分辨率分析及识别研究[D].哈尔滨:哈尔滨工程大学,2009.
[3] FEI T, KRAUS D, ZOUBIR A M. Contributions to automatic target recognition systems for underwater mine classification[J]. IEEE Transactions on Geoscience & Remote Sensing, 2014,53(1):505-518.
[4] 石敏,徐袭.小波包熵在水下目标识别中的应用研究[J].计算机工程与应用,2014,50(1):215-217.
[5] LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015(2):436-444.
[6] SUN Y, WANG X, TANG X. Deep learning face representation from predicting 10,000 classes[C]. IEEE Conference on Computer Vision and Pattern Recognition, Washington D C,2014.
[7] HUANG G, LIU Z, MAATEN L V D, et al. Densely connected convolutional networks[J]. Nature, 2016(1):55-59.
[8] QIN H, LI X, LIANG J, et al. Deep fish: accurate underwater live fish recognition with a deep architecture[J]. Neurocomputing, 2016(1):49-58.
[9] SRIVASTAVA N, HINTON G, KRIZHEVSKY A, et al. Dropout: a simple way to prevent neural networks from overfitting[J]. Journal of Machine Learning Research, 2014,15(1):1929-1958.
[10] ATAMAN E, AATRE V K, WONG K M. Some statistical properties of median filters[J]. IEEE Transactions on Acoustics Speech & Signal Processing, 1981,29(5):1073-1075.
[11] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Imagenet classification with deep convolutional neural networks[C]. International Conference on Neural Information Processing Systems, London, 2012.
[12] HINTON G E, SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006(2):504-507.
[13] DENG J, DONG W, SOCHER R, et al. Imagenet : a large-scale hierarchical image database[C].2009 IEEE Conference on Computer Vision and Pattern Recognition, CVPR2009, Miami, 2009.
[14] BOOM B J, HUANG P X, HE J, et al. Supporting ground-truth annotation of image datasets using clustering[C].21st International Conference on Pattern Recognition, Tsukuba, 2012.
[15] JIA Y Q, SHELHAMER L. Caffe: convolutional architecture for fast feature embedding[J]. Eprint Arxiv, 2014(2):675-678.
