In the embedded system environment, both large amount of face image data and the slow recognition process speed are the main problem facing face recognition of end devices. This paper proposes a face recognition algorithm based on dual-channel images and adopts a cropped VGG-like model referred as VGG-cut model for predicting. The training set uses the same single-layer images of the same person combined into dual channels as a positive example, and single-layer images of different people are combined into dual channels as a negative example. After model training, the trained face verification model is used as the basis of face recognition, and the final recognition result can be obtained through loop matching and similarity ranking. The experimental results on a RISC-V embedded FPGA platform show that compared with the ResNet model called in Dlib, the VGG-like model and MobileFaceNet trained by Keras, our algorithm is increased by 240×, 88×, and 19× in recognition speed, respectively without significant accuracy reduction.

Gastric intestinal metaplasia (GIM) is regarded as a remarkable precursor
for the development of intestinal-type stomach cancer. Goblet cell (GC) segmentation
is the crucial step for assessing the degree of GIM by confocal laser endomicroscopy
(CLE). However, GC segmentation by hand is difficult, unreliable, and time consuming. Meanwhile, due to the high resolution and noise interference of CLE
images, existing segmentation approaches perform poorly on this task. To tackle those
issue, we collected 343 confocal laser endomicroscopy images of 62 patients from a
Grade-A tertiary hospital. Each CLE image is manually annotated and then verified
three times by skilled medical specialists. Then, U-Net is improved by incorporating
the pixel gradient attention mechanism, which focuses on color gradient information
around GC and captures color gradient features to direct feature maps in the skip
connection layer. At last, the model output vector is used to calculate the possibility
map and generate the final segmentation area. Compared with mainstream models,
GCSCLE performs the best segmentation result when tested on our CLE dataset and
achieved an IOU of 87.95% and a DICE of 86.64%. Our result shows, the performance
of the GCSCLE can be compared with the manual CLE image processing in clinical
settings, and it can improve segmentation accuracy and save time and costs.

Resolution Band Width (RBW) represents the 3 dB bandwidth of the internal IF filter of the spectrum analyzer, the setting of which greatly affects the clarity of the output spectrum line of the spectrum analyzer. Its reasonable value is essential to correctly distinguish the signals of different frequencies, especially in the case of aliasing of the output spectrum. Thus, it is of great significance to study the relationship between the setting of RBW and the output spectrum. Due to the expensive spectrum analyzer, the RBW adjustment range, frequency range and output accuracy are limited, so this paper will apply an economic, RBW arbitrary adjustable, high accuracy equivalent convolution method for the simulation of the RBW parameters of the spectrum analyzer, and evaluate the effect of RBW and filter type on the degree of output spectrum aliasing based on a typical equal-amplitude two-tone signal input, and finally give setting suggestions on RBW and filter type.