foreword
Friends who are familiar with me should know that I participated in many competitions in college. I have always been more enthusiastic about competitions. From my personal point of view, I think competitions can stimulate learning enthusiasm and test their own skills. College students are very There are few opportunities to understand the needs of the market outside the classroom. The outsourcing service competition is a good opportunity to help college students get in touch with the services and talent needs of the market. In addition, many competitions in universities can accumulate their own conduct points and get Getting a good ranking has gained a lot. It can be said that academic career kills two birds with one stone. It just so happens that the 14th China College Student Service Outsourcing Competition has come to a successful conclusion. I have also participated in such competitions and won gold medals in my college competition career. I just take this opportunity to take a closer look at the China College Student Service Outsourcing Competition.
game background
The China University Student Service Outsourcing Innovation and Entrepreneurship Competition (hereinafter referred to as the "Competition") is an annual national competition held in response to the country's relevant strategic measures and calls for encouraging the development of the service outsourcing industry and strengthening the training of service outsourcing talents. Service outsourcing refers to the outsourcing of certain business or work process to other institutions or companies to complete, to a certain extent reduce the cost of enterprises and improve efficiency. Winning the competition award for service outsourcing can prove that the winner has strong ability and experience in teamwork, project management, communication and coordination, etc., which will be helpful for future employment and personal career development. Moreover, this kind of competition generally has research bonus points in major colleges and universities, so naturally more people pay attention.
Since its establishment in 2010, the competition has attracted more than 1,600 colleges and universities and more than 500,000 college students to participate, and has become a national competition in the field of service outsourcing industry. A total of 8006 teams from 803 colleges and universities across the country signed up for this competition, and the number of teams registered reached a new high. Moreover, with the continuous iteration of technology, the market demand is also constantly changing. The competition theme of the competition also responds to the development of the times, and many questions that are appropriate to the theme of cutting-edge technology have been born. For example, with regard to the recently popular application of AI large models, this competition has well combined the current hot technical topics and inspired more college students participating in the competition to think.
The content design of this competition fully focuses on the practical problems such as technology and management faced in the development of enterprises, and is more closely integrated with the industry. Intelligent character recognition technology is one of the technologies that the competition focuses on. Intelligent text recognition technology integrates intelligent image processing, optical character recognition, deep learning, natural language processing and other technologies, and can recognize, analyze and understand text information in complex scenarios such as multi-language, multi-format, curved surface, folds, background interference, etc. It has a wide range of applications in many fields such as production, education, and life.
Hehe Information has more than ten years of in-depth experience in the field of intelligent text recognition. Based on its own cognition in the industry, starting from issues such as "innovative application of personal financial management based on intelligent text scenarios", it issued a "product collection order" to the majority of students. ", received positive feedback, and related questions attracted nearly 300 teams from more than 70 universities across the country, including Chongqing University, Beijing University of Posts and Telecommunications, Harbin Institute of Technology, East China University of Science and Technology, Sichuan University, and Northwest University. Produced many excellent works.
Comments on entries
In this competition, many college teams have a good insight into the pain points of different user groups. For example, from the phenomenon that "middle-aged and elderly people need to go through cumbersome operations when using bookkeeping applications", they found that "picture recognition bookkeeping is more popular." trends, and optimize product development in a targeted manner.
Take Central South University's Almighty King of Bookkeeping - OCR bill recognition system project based on image preprocessing and BERT model as an example.
Project Design Direction
In response to this demand, Central South University has a good understanding of the difficulties in project development. First of all, this type of application is aimed at middle-aged and elderly people, so it is more appropriate to develop Android phone APKs suitable for middle-aged and elderly people. In addition, there is no need to design more complicated functions, and it is only suitable for bookkeeping. The application should be simple and clear in operation, easy to get started quickly, and it is best not to design too many complicated processes. Traditional bookkeeping is generally manually entered or other application-assisted account entry, which is cumbersome. In this competition, with the help of Hehe Information's intelligent image recognition module, small ticket recognition can be performed quickly, and it can be read directly into the corresponding frame content according to the template. Therefore, how to combine the intelligent image recognition module and design a simple and efficient data input financial management tool is the main problem of the project.
The work of Central South University adopts the SpringCloud micro-service architecture. Spring Cloud can help developers realize the rapid construction and deployment of the micro-service architecture, and improve the scalability and elasticity of the system. For the competition rapid development service, there is no problem in choosing this architecture, but how to perform image recognition and the technical requirements for recognition accuracy are relatively difficult. OCR recognition is a good technology for technology stacks that can support the recognition of takeaway bills, supermarket receipts, hand accounts, securities, Alipay WeChat consumption screenshots, etc., and have a high accuracy rate. The corresponding classification model and information extraction labeling model. The above technical difficulties have been taken into account by Central South University and have been dealt with in a relatively standard way, such as Python's jieba word segmentation library and the text conversion technology Word2Vec, information extraction and relationship extraction have taken into account the sequence labeling model (BiLSTM-CRF). The above technologies are relatively mature, there will be no major problems in use, and they are suitable for rapid development. In addition, the work also takes into account the enterprise side, and makes another technical framework that can obtain consumers' consumption habits, place advertisements, and improve bill recognition, which is relatively comprehensive.
In addition, if it is possible to automatically generate historical bill portraits based on existing historical data, or to automatically evaluate account data in combination with AI large models, it may be more comprehensive. Even so, the degree of completion of this work is high enough, and it is simple, efficient and most suitable for the theme of the competition.
Development direction
The code development direction of Central South University's works is very clear. The front end uses the Flutter framework, and Flutter uses a self-drawing engine with excellent performance and rendering effects. By using a rich widget library, developers can easily create beautiful user interfaces and achieve a high degree of customization. This makes Flutter competitive in apps with a strong sense of design, branded apps, and projects that focus on user experience. The development mode adopts the mode of front-end and back-end separation. In this mode, the front-end and back-end can be developed independently without interfering with each other and cooperate with each other. Finally, they are combined into a complete application program, which improves the development efficiency and system maintainability. and scalability, the front-end and back-end interaction is simple and clear, and it is easy to debug and test.
The Almighty King of Accounting Works by Central South University—the OCR bill recognition system based on image preprocessing and BERT model not only solves the pain points of financial receipts in the market, but also flexibly combines Hehe Information's intelligent image recognition module with the recognition of receipts API implements financial management scenarios, converts complex receipt information into simple and efficient data input in seconds, and builds a financial management tool in multiple scenarios, allowing users to easily perform bookkeeping and financial management in different scenarios .
It is no exaggeration to say that BERT has revolutionized natural language processing significantly. For example, using a model trained on a large unlabeled data set, it achieved excellent results in 11 independent natural language processing tasks. It can be achieved with a little fine-tuning. BERT has triggered many new natural language processing architectures, training methods, and language models, such as Google TransformerXL, OpenAI's GPT-2, XLNeT, ERNIE2.0, RoBERTa, etc. One of the biggest challenges in natural language processing is the lack of training data. Because natural language processing is a diverse field with many different tasks, most task-specific datasets contain only a few thousand or hundreds of thousands of human-labeled training examples. Therefore, being able to support a huge data set is the key to combining large models with other fields. Running BERT is a GPU-intensive task that requires a large amount of computing resources to support.
When it comes to image processing technology, we have to talk about OCR technology. With the continuous development of technology, the application of OCR technology in the industry has also expanded from the initial simple optical character recognition to the field of intelligent document processing covering image preprocessing, text recognition, layout analysis, document understanding and other technologies. The combination of document processing and artificial intelligence can realize the automatic processing and analysis of a large number of documents, improve work efficiency and accuracy, and reduce labor and time costs, which is of great significance to the knowledge management and business analysis of enterprises. Relying on the self-developed intelligent text recognition service platform, Hehe Information has made some relevant work progress and research results in solving various problems faced by the industry, and discussed some key technical problems and challenges currently faced by the industry. I believe that Hehe Information has accumulated a lot in the fields of pattern recognition, deep learning, image processing, and natural language processing, and will benefit more people with technical solutions.
Hehe Information's intelligent document processing technology adopts technologies such as precise image cropping, deformation correction, and shadow and moiré removal, and uses artificial intelligence technology to enhance and improve the clarity of document images, thereby improving the quality of document images and reading experience. In this way, the quality and efficiency of downstream tasks in document processing, such as recognition conversion and image analysis, can be effectively improved. At present, this technology has been applied to intelligent text recognition products, providing services to hundreds of millions of users from hundreds of countries and regions around the world.
The combination of document processing and artificial intelligence can realize the automatic processing and analysis of a large number of documents, improve work efficiency and accuracy, and reduce labor and time costs, which is of great significance to the knowledge management and business analysis of enterprises. In addition, the future graphic and text intelligent processing technology will also be more customizable, providing customers with tailor-made solutions according to different industries and application scenarios. This will help meet the different needs of customers and enhance customer experience and satisfaction.
game development
From the perspective of technological innovation, the participating teams showed profound thinking. They blend advanced large-model technology with real-world needs in the field of intelligent document processing. By applying text recognition and text understanding techniques to a large volume of consumer receipts, they succeeded in achieving efficient classification. This innovation has transformed the financial close process from a tedious manual process into a highly automated process. This practice fully demonstrates their ability to apply new technologies to solve traditional problems.
According to big data statistics, more than 70% of college students are engaged in occupations that have nothing to do with their undergraduate majors after graduation. This means that when college students leave the university, they may be greeted by a completely unfamiliar field. If they do not want to be eliminated, they must constantly enrich and improve their knowledge reserves. It can be seen that it is very important to establish a diversified talent training mechanism. At present, the innovation, breakthrough and development of science and technology are more and more dependent on the intersection and integration of multiple disciplines, which puts forward higher requirements for the cultivation of interdisciplinary talents. The competition's scoring criteria for entries is also very "simulated", involving technical resources and economic cost control, judgment of project creative prospects, analysis of market demand, etc., covering commercial value, social application value and other aspects of evaluation.
In this era of artificial intelligence, technology companies play a key role in application scenarios and play an important role in talent cultivation. In the future, all walks of life are eager to have talents who have a deep understanding of professional academic fields and have the potential to solve problems, especially in the growing field of artificial intelligence. In this context, the competition provides students with an excellent display platform in the field of science and technology.
By participating in such activities, technology companies can not only tap potential talents, but also establish closer ties with the academic community to jointly promote technological development and innovation. In short, this kind of school-enterprise cooperation has made remarkable achievements in personnel training and industry development, laying a solid foundation for future scientific and technological progress.