论文部分内容阅读
针对海面背景舰船目标单一波段图像识别率低的问题,提出了一种基于卷积神经网络(CNN)的融合识别方法。该方法提取可见光、中波红外和长波红外3个波段舰船目标特征进行融合识别。模型主要分为3个步骤:通过设计的6层CNN,同时对三波段图像进行特征提取;利用基于互信息的特征选择方法对串联的三波段特征向量按照重要性进行排序,并按照图像清晰度评价指标选取固定长度的特征向量作为目标识别依据;通过额外的2个全连接层和输出层进行回归训练。采用自建的三波段舰船图像数据库进行模型的训练和测试,共包含6类目标,5000余张图像。实验结果表明,本文方法识别率达到84.5%,与单波段识别方法相比有明显提升。
Aiming at the low recognition rate of a single band image of a ship target in a sea background, a fusion identification method based on convolution neural network (CNN) is proposed. This method extracts the target features of three bands of visible light, medium-wave infrared and long-wave infrared for the fusion recognition. The model is mainly divided into three steps: through the design of the 6-layer CNN, while the three-band image feature extraction; the use of mutual information based feature selection method of the series of three-band eigenvector in accordance with the importance of sorting, and in accordance with the image sharpness The evaluation index selects fixed-length eigenvectors as the basis for target recognition; regression training is performed through an extra two fully connected layers and output layers. The self-built three-band ship image database is used to train and test the model, which contains 6 types of targets and more than 5,000 images. The experimental results show that the recognition rate of this method reaches 84.5%, which is obviously improved compared with the single-band identification method.