High quality medical images are not only an important basis for doctors to carry out clinical diagnosis and treatment, but also conducive to downstream tasks such as image analysis. Although many medical image enhancement methods have achieved good results, some of them still have shortcomings in homogenizing illumination distribution and maintaining texture details, and even introduce boundary artifact noise. In order to deal with these problems, this paper proposes a multi-scale attention generative adversarial network (MAGAN) for medical image enhancement, which is suitable for unpaired images. Our MAGAN is trained in the confrontation between two generators and two discriminators. It tries to fuse multi-scale information in feature extraction by establishing feature pyramid, and filters irrelevant activation to highlight important regions based on attention distribution, which is positive for imaging. Moreover, MAGAN strengthens the constraints on the quality of enhanced image from the perspectives of illumination distribution, texture details, deep semantic features and smoothness, so as to improve the enhancement effect. Compared with six state-of-the-art methods, the experimental results show that MAGAN has the most significant image enhancement effect, and also performs best in the downstream task of image segmentation.

For a broad range of applications, hyperspectral image (HSI) classification is a hot topic in remote sensing, and convolutional neural network (CNN)-based methods are drawing increasing attention. However, to train millions of parameters in CNN requires a large number of labeled training samples, which are difficult to collect. A conventional Gabor filter can effectively extract spatial information with different scales and orientations without training, but it may be missing some important discriminative information. In this article, we propose the Gabor ensemble filter (GEF), a new convolutional filter to extract deep features for HSI with fewer trainable parameters. GEF filters each input channel by some fixed Gabor filters and learnable filters simultaneously, then reduces the dimensions by some learnable $1\times 1$ filters to generate the output channels. The fixed Gabor filters can extract common features with different scales and orientations, while the learnable filters can learn some complementary features that Gabor filters cannot extract. Based on GEF, we design a network architecture for HSI classification, which extracts deep features and can learn from limited training samples. In order to simultaneously learn more discriminative features and an end-to-end system, we propose to introduce the local discriminant structure for cross-entropy loss by combining the triplet hard loss. Results of experiments on three HSI datasets show that the proposed method has significantly higher classification accuracy than other state-of-the-art methods. Moreover, the proposed method is speedy for both training and testing.