Lianjia.Com Transaction Data Crawler

Remote Internship Program of Beijing Normal University [Link]

• Implemented a house transaction data crawler by using sniffer tool and Requests/BeautifulSoup/PyQuery/XPath library to parse the webpage, using MongoDB to store collected data and implementing crawler parallelization based on multi-threading.
• Cracked the anti-crawler strategy by decompiling Android APP for simulating login.
• Implemented a dynamic IP proxy pool based on Redis and Flask for reuse third-party agent IP which can improve agent IP availability and crack anti-crawler strategy.
• Crawled 1.05 million housing transaction data in 16 districts of Beijing city.

Autonomous navigation robots with integrated multi-sensors

• Built a multi-sensor integrated indoor navigation robot for the blind based on Linux and ROS; Implemented indoor mapping, GPS-free auto-positioning, path planning, obstacle detection, and speech recognition control function; Improved the position deviation accuracy less than 5 cm.
• Optimized the traditional mapping method, proposed a hybrid method that can simplify mapping processing and improve usability; Implemented a Deep Learning based real-time obstacle detection method by using Intel Neural Compute Stick with parallel acceleration into the robot; The speed of 4-stick parallel detection was 24 FPS with the power consumption reducing by 7 times.

Modern Poetry Generator

2018 JD.com AI Camp Project (Top 1 of NLP Teams) [Code]

• Implemented a retrieval-based method by using jieba, adopting Gensim TF-IDF to calculate text similarity, and increasing the history weighting strategy to improve the poetry coherence.
• Implemented a generation-based model by using TextRank and Word2Vec to rank the keywords, applying Bi-RNN and Attention mechanism to train and generate poetry.
• Crawled 20,000 modern poetries as an expanded corpus to project.

Devil Self-Driving Micro Car

2018 Sinovation Ventures DeeCamp AI Camp project (Best Team in 2018) [Code]

• Implemented a self-driving car based on ROS and multi-sensor, including the units of sensing, computation, and control.
• Used Tensor RT to speed up the process of deep learning based on TensorFlow.
• Used the CARLA simulator to simulate the environment in the driving process.

Morphological Features Detection System Of Trachitus Ovatus Fish

Internship Project of Hainan University

• Designed a system to detect fish morphological features; Implemented the analysis module by C++ based on OpenCV; Implemented GUI interface for visualization by Qt framework. Proposed a fishtail contour fit method by image processing and curve fitting; Implemented a Machine Learning method Adaboost for locating and recognizing eye features. Three theses and two patents were published based on this project.
• Implemented a Machine Learning method Adaboost for locating and recognizing eye features; Optimized two fit methods for recognizing eye features. The speed of fishtail detection and fisheye feature detection increasing by 3 times and 12 times respectively.