Code review is an important step in developing embedded software for aerospace. In recent years, with the increase of requirements in embedded software development, the traditional manualbased code review method has gradually failed to meet the fastpaced software development. In order to relieve the pressure of reviewer and prompt the efficient of code review, the researchers turn their attention to the field of artificial intelligence, especially the deep learning technology, and make efforts to find technological breakthroughs so as to develop a more intelligent review system. This article summarizes the research work related to the application of artificial intelligence technology in code review and code comprehension in recent years. First, the difficulties of embedded code review and the solution are briefly described. Furthermore, the key activity, program comprehension, is introduced. And the application of program comprehension based on artificial intelligence in code review is explained in detail. Finally, according to the development of related fields, the future tendency of code review is further discussed.

Aiming at the problem that the activation function in Convolutional Neural Networks can’t effectively provide a specific gradient response for pixels at different activation levels, an adaptive activation function is designed, which is composed of multiple piecewise linear functions. Firstly, according to the ranges of pixel activation value, multiple independent activation domains are adaptively generated, and the union of each activation domain contains the activation values of all pixels in the activation map; then a specific linear function is learned in each activation domain in order to provide a specific gradient response for the pixels in the activation domain; finally taking the NIN and ResNet18 as examples, on the CIFAR10 and CIFAR100 dataset, the performance of the proposed activation function is verified. Experimental results show that, compared with the existing activation function, the activation function proposed in this paper can provide a suitable gradient response for pixels with different activation levels, so that the classification accuracies reach 87.96% and 69.01% on the NIN respectively. The accuracies reach 88.56% and 73.54% on the ResNet18, respectively.

 In recent years, the requirements of hardware computing power for neural network algorithms that showing extremely superior performance have gradually become higher. In some lowpower scenarios such as spaceborne systems, FPGAs with low power consumption and high parallelism are the most suitable hardware acceleration platforms for neural network algorithms. In order to solve the problems of high onchip resource consumption and high coupling of various operation modules in hardware structure design, a set of dedicated instruction set is designed and implemented to the structural design of FPGAbased neural network accelerator. Firstly, the design and application of the instruction set are introduced. The whole system is composed of instruction register, instruction interpreter, instruction forwarding module, memory management unit, and multiple execution modules. The system can realize the multiplexing of different operation modules and reduce the coupling between modules. Afterwards, the YOLOV3Tiny network model is used as an example to compare the onchip resource consumption of two acceleration schemes, tiled and commandcontrolled. It is verified that the application of dedicated instruction set can effectively reduce the use of FPGA onchip resources.
Keywords： command set；neural network；FPGA

Satellite Internet is becoming a kind of space infrastructure construction. The contradiction between the demand of largescale satellite launch and the exponential growth of ground operation and control management tasks is increasingly prominent. In this paper, through the architecture design of parameter binding, logic sublayer and data interaction module, a framework of autonomous management software system based on model configuration is proposed. Aiming at the core logic sublayer, the active remote command model, telemetry active model and data fusion interactive control model are established to realize the rapid development of autonomous management software for satellite, the problem of continuous function expansion in design and test verification is discussed. Finally, the design of satellite FDIR autonomous management function is verified. The results show that the software system design framework can meet the requirements of satellite autonomous control function design, improve the efficiency of software development, and further improve the intelligence level of satellite lay the foundation.

 Each behavior of spacecraft needs a variety of information synthesis to make a judgment. How to effectively transfer the data of multiple sensors has become a concern for many people. The robot operating system used a topic based method to solve this problem. However, ROS is written by C++ language and can’t be directly used in embedded platform. In this paper, we regard information transference as nodes, and use C language to implement the communication mechanism. Each node can choose the topic interest to become the publisher or subscriber. The publisher and subscriber do not know the information of other nodes to realize function decoupling. AT present, the method achieves perfect result between single and multimachine on windows platform.

A method for routing data is proposed in this paper for satellite constellation network. The method combine the widely used OSPF dynamic routing strategy in ground networks and the traditional beforehand computed satellite constellation static routing strategy. Since the dynamic routing strategies can be more reliable and robust but consume more computing resources and network resources, OSPF routing strategy is compressed to meet the hardware requirements in satellite constellation network. Through automatically switch between static and dynamic routing strategy, the satellite constellation network can obtain more reliabilities.