Name:Platform user evaluation opinion extraction
Key objectives:Analyze and judge emotions for massive user evaluation data
Description:
Firstly, according to the user's comment data, the user's evaluation and impression of the platform are extracted. The influence of external events on the user's commentary is analyzed. The time is oriented, then the comment is de-emphasized, and then the comment is divided into Chinese and part-of-speech. Measure the importance of words in a document, extract dependencies, and derive labels. Finally, the commentary sentiment analysis is carried out. When the sentiment analysis of the data is performed, we use snownlp to collect, compare, and train the software-related training sets. After the sentiment analysis of the data, generate satisfaction indicators and display them graphically
Name:Image Caption
Key objectives:Describe the content of a given image
Description:
First, the data is processed, the image features are extracted through the neural network, the image is preprocessed with ResNet, and the output of the specified layer is obtained and saved. The description generates a word vector by segmentation, filtering, pad, and then uses the LSTM output to fully predict the words appearing below and calculate the loss, using cross entropy as the objective function. After training, test the picture, extract features from the picture, and use LSTM to get the output prediction. After this step, embedding to get the word vector and then input LSTM again and repeat until the end of the generated text description.
Name:Face sign-in system
Key objectives:First enter the personnel information in the system, the relevant personnel check in at the front of the computer, display the name and welcome information of the sign-in personnel in real time, and finally display the list of unsigned persons.
Description:
A face recognition sign-in program was prepared for the visit.
Personnel recognition success rate is 100%. The code program was written by myself independently. The entire system is well packaged, which simplifies the user's use to the greatest extent and is robust enough to be used in a variety of applications.
The organizer only needs to save the photo file to the specified location according to the format requirements, and the system can identify and process it completely.
When the system is running, the name of the person who is signed in can be displayed in real time, and the welcome message is displayed. The program finally displays a list of unsigned people.
Easy to use and easy to operate. The system writing structure is clear and efficient. A good practical use effect was achieved in the event.
Year:2016
Name:Research on gait planning and obstacle avoidance of biped robot (National Undergraduate Innovation Project)
Awards:
published the paper 《A Novel Fuzzy Omni-directional the Gait Planning Algorithm for Biped Robot》
2016 The eighteenth national robot championship and the Seventh International Robot Olympic Champion(Best ever in school history)
2016 RoboCup Robot World Cup China runner-up (Best ever in school history)
Description：
Firstly humanoid robot modeling, Mechanical analysis and modeling,gait planning modeling in the simulation platform,and the MATLAB simulation is used to simulate the prediction of multiple obstacles.The factors that influence the stability of the robot under different material and blocking conditions are reasonably deduced.Under the condition of reasonable to infer the robot in different materials, the influence of road congestion under the impact factor of the gait stability, combined with the robot field testing. During the study period, I published the paper 《A Novel Fuzzy Omni-directional the Gait Planning Algorithm for Biped Robot》
Year:2017
Name:Research on swarm intelligence cooperation technology of NAO robot (National Undergraduate Innovation Project)
Awards: patent《Robot Network Communicator》.
2017 Robot Innovation Challenge Northwest Division Silver
Description:
In order to overcome the situation that robot just communicate with the host,which always cause the single point of failure and the island effect.After trying a variety of communication channels, such as Bluetooth, WiFi, and finally established the WiFi udp/tcp hybrid switching mode.The communication between the robot to break the barriers, making a robot vision,positioning and other resources can be shared by the whole.Finally can reach the skill points, in the host mode,resource can be collected by each one and after host intelligent decision to redistribution.In the no host mode,each robot pass message by ask/answer mechanism.During the study, I published the patent《Robot Network Communicator》
Year:2019
Name:Image Caption
Description:
First, the data is processed, the image features are extracted through the neural network, the image is preprocessed with ResNet, and the output of the specified layer is obtained and saved. The description generates a word vector by segmentation, filtering, pad, and then uses the LSTM output to fully predict the words appearing below and calculate the loss, using cross entropy as the objective function. After training, test the picture, extract features from the picture, and use LSTM to get the output prediction. After this step, embedding to get the word vector and then input LSTM again and repeat until the end of the generated text description.
Year:2019
Name:Face sign-in system
Awards: The actual field test success rate of the program is 100%
Description:
Face sign-in system
A face recognition sign-in program was prepared for the visit.
Personnel recognition success rate is 100%. The code program was written by myself independently. The entire system is well packaged, which simplifies the user's use to the greatest extent and is robust enough to be used in a variety of applications.
The organizer only needs to save the photo file to the specified location according to the format requirements, and the system can identify and process it completely.
When the system is running, the name of the person who is signed in can be displayed in real time, and the welcome message is displayed. The program finally displays a list of unsigned people.
Easy to use and easy to operate. The system writing structure is clear and efficient. A good practical use effect was achieved in the event.