Convolutional Neural Network with Spatial-Variant Convolution Kernel

Abstract

Radar images suffer from the impact of sidelobes. Several sidelobe-suppressing methods including the convolutional neural network (CNN)-based one has been proposed. However, the point spread function (PSF) in the radar images is sometimes spatially variant and affects the performance of the CNN. We propose the spatial-variant convolutional neural network (SV-CNN) aimed at this problem. It will also perform well in other conditions when there are spatially variant features. The convolutional kernels of the CNN can detect motifs with some distinctive features and are invariant to the local position of the motifs. This makes the convolutional neural networks widely used in image processing fields such as image recognition, handwriting recognition, image super-resolution, and semantic segmentation. They also perform well in radar image enhancement. However, the local position invariant character might not be good for radar image enhancement, when features of motifs (also known as the point spread function in the radar imaging field) vary with the positions. In this paper, we proposed an SV-CNN with spatial-variant convolution kernels (SV-CK). Its function is illustrated through a special application of enhancing the radar images. After being trained using radar images with position-codings as the samples, the SV-CNN can enhance the radar images. Because the SV-CNN reads information of the local position contained in the position-coding, it performs better than the conventional CNN. The advance of the proposed SV-CNN is tested using both simulated and real radar images.

Image semantic segmentation method based on GAN network and ENet model

“…As a result, super-resolution images with similar distribution to high-resolution images can be generated. Super-resolution generative confrontation network is a method of using residual structure in the generator [32]. It adds a residual structure after the convolutional layer of the convolutional neural network, which makes the network have a deeper layer and adds non-linear features, thereby greatly improving the learning ability of the model.…”

Editorial for the Special Issue “Advanced Techniques for Ground Penetrating Radar Imaging”

“…The latter is of special interest when the surveyed volume is much greater than the targets and features, so visual inspection of the volume is extremely time-consuming and challenging. Besides, machine learning techniques have been also introduced to improve GPR image quality, e.g., by means of sidelobe suppression [9].…”

“…The latter can be affected by the fact that the Point Spread Function (PSF) in the radar images can be sometimes spatially variant. A Spatial-Variant CNN (SV-CNN) with spatial-variant convolutional kernels is proposed in [9] to overcome this issue, proving its better performance compared to the conventional CNN in realistic scenarios.…”

“…The scope of the Special Issue contributions can be classified into three main groups: High-resolution GPR systems [1][2][3][4], noise mitigation in GPR measurements [5,6], and GPR data processing enhancement [7][8][9][10].…”

Blind Non-Uniform Motion Deblurring using Atrous Spatial Pyramid Deformable Convolution and Deblurring-Reblurring Consistency

“…The pixel-adaptive convolution of Su et al [47] can be regarded as a simplified version of the paired convolution [9] by reducing the kernel size to one. Dai et al [7] exploit the position information of each pixel. The local patches extracted by the data kernel are concatenated with their positions in the image, and then are passed to a convolution layer.…”