The fusion of infrared and visible images can utilize the indication characteristics and the textural details of source images to realize the all-weather detection. The deep learning (DL) based fusion solutions can reduce the computational cost and complexity compared with traditional methods since there is no need to design complex feature extraction methods and fusion rules. There are no standard reference images and the publicly available infrared and visible image pairs are scarce. Most supervised DL-based solutions have to take pre-training on other labeled large datasets which may not behave well when testing. The few unsupervised fusion methods can hardly obtain ideal images with good visual impression. In this paper, an infrared and visible image fusion method based on unsupervised convolutional neural network is proposed. When designing the network structure, densely connected convolutional network (DenseNet) is used as the sub-network for feature extraction and reconstruction to ensure that more information of source images can be retained in the fusion images. As to loss function, the perceptual loss is creatively introduced and combined with the structure similarity loss to constrain the updating of weight parameters during the back propagation. The perceptual loss designed helps to improve the visual information fidelity (VIF) of the fusion image effectively. Experimental results show that this method can obtain fusion images with prominent targets and obvious details. Compared with other 7 traditional and deep learning methods, the fusion results of this method are better on objective evaluation and visual observation when taken together.INDEX TERMS Infrared and visible images, deep learning, unsupervised image fusion, densely connected convolutional network, perceptual loss.

Diagnosing incipient faults of rotating machines is very important for reducing economic losses and avoiding accidents caused by faults. However, diagnoses of locations and sizes of incipient faults are very difficult in a noisy background. In this paper, we propose a fault diagnosis method that combines kernel principal component analysis (KPCA) and deep belief network (DBN) to detect sizes and locations of incipient faults on rolling bearings. Effective information of raw vibration signals processed by KPCA method is used as input signals of the DBN of which weights of the first RBM are initialized by contribution rates of principal components. A DBN with complex structures can be cut into a briefer network by KPCA-DBN model. That model reduces network structure and increases convergence rate. As a result, an average test accuracy by KPCA-DBN can reach 99.1% for identification of 12 labels including incipient faults and the training time is 28s which is half of that by DBN model. The average accuracy of rolling bearing location detection nearly gets to 100% and the average accuracy of fault size detection is above 99%. Compared with SVM, BP, CNN, Deep EMD-PCA, CNN-SVM and DBN, it is found that training time can be shortened and detection accuracy can be improved by KPCA-DBN model. The proposed method is beneficial to realize sizes and locations detection of incipient faults online.

With the extensive coverage of the rail transit system, ensuring the safe operation of rail vehicles is an important prerequisite. Insufficient lubrication will cause friction and wear of axle box bearings, which is directly related to ensured safety of high-speed trains. A non-Newtonian elastohydrodynamic lubrication(EHL) between tapered rolling elements and inner ring of axle box bearing in high-speed trains was established by numeric simulation. The input parameters of working conditions, including velocity, acceleration and plastic viscosity, were changed, considering the actual application and their influence trends on film-forming characteristics were analyzed. As a result, a phase of acceleration of starting or a process of braking at a low speed tends to occur mixed lubrication. Therefore, a method of optimizing surface morphology of rolling elements was adopted to improve lubrication. Based on comparison experiments, it was recommended that RMS roughness was greater than 0.03 μm and less than 0.1 μm and kurtosis was three and skewness was negative in a range of −1 to −0.5 and texture direction was parallel to rotation direction. The optimized surface promotes the transition from mixed-lubrication to full film lubrication, which alleviated the problem of surface damage due to insufficient lubrication and prolongated the service life.