Metal oxide gas sensors are predominant solid-state gas detecting devices for domestic, commercial and industrial applications, which have many advantages such as low cost, easy production, and compact size. However, the performance of such sensors is significantly influenced by the morphology and structure of sensing materials, resulting in a great obstacle for gas sensors based on bulk materials or dense films to achieve highly-sensitive properties. Lots of metal oxide nanostructures have been developed to improve the gas sensing properties such as sensitivity, selectivity, response speed, and so on. Here, we provide a brief overview of metal oxide nanostructures and their gas sensing properties from the aspects of particle size, morphology and doping. When the particle size of metal oxide is close to or less than double thickness of the space-charge layer, the sensitivity of the sensor will increase remarkably, which would be called “small size effect”, yet small size of metal oxide nanoparticles will be compactly sintered together during the film coating process which is disadvantage for gas diffusion in them. In view of those reasons, nanostructures with many kinds of shapes such as porous nanotubes, porous nanospheres and so on have been investigated, that not only possessed large surface area and relatively mass reactive sites, but also formed relatively loose film structures which is an advantage for gas diffusion. Besides, doping is also an effective method to decrease particle size and improve gas sensing properties. Therefore, the gas sensing properties of metal oxide nanostructures assembled by nanoparticles are reviewed in this article. The effect of doping is also summarized and finally the perspectives of metal oxide gas sensor are given.

Aiming at the limitation of the convolution kernel with a fixed receptive field and unknown prior to optimal network width in U-Net, multi-scale U-Net (MSU-Net) is proposed by us for medical image segmentation. First, multiple convolution sequence is used to extract more semantic features from the images. Second, the convolution kernel with different receptive fields is used to make features more diverse. The problem of unknown network width is alleviated by efficient integration of convolution kernel with different receptive fields. In addition, the multi-scale block is extended to other variants of the original U-Net to verify its universality. Five different medical image segmentation datasets are used to evaluate MSU-Net. A variety of imaging modalities are included in these datasets, such as electron microscopy, dermoscope, ultrasound, etc. Intersection over Union (IoU) of MSU-Net on each dataset are 0.771, 0.867, 0.708, 0.900, and 0.702, respectively. Experimental results show that MSU-Net achieves the best performance on different datasets. Our implementation is available athttps://github.com/CN-zdy/MSU_Net.