Project Background
Insects, as a treasure trove of biological diversity, are facing multiple challenges such as a lack of understanding, a lack of conservation awareness, and excessive human activity. These factors have collectively led to adramatic decline in insect populations. This decline not only threatens the balance of ecosystems but also impacts humanity’s ability to obtain resources and benefits from nature. Especially in theagriculture and forestry sectors, traditional manual identification methods have become inadequate due to inefficiency and high costs, particularly in theprevention and control of crop pests and monitoring of forest insects.
However, with the rapid advancement of artificial intelligence and machine learning technologies, especially the significant breakthroughs of deep learning algorithms in image recognition, we are ushering in a new opportunity for insect recognition. Against this backdrop, the “Insect Recognition System Based on CNN” project team at Beijing Forestry University has utilized advanced deep learning models and algorithms to construct a more comprehensiveintelligent insect recognition system. This system can achieveefficient, fast, and accurate recognition of insect species, providing technical support to alleviate the aforementioned issues and promote the sustainable development of agriculture and forestry.
Project Introduction
“Smart Eye for Insects, Intelligent Protection of Ecology, Technology Empowerment, Foundation of Agriculture.”This project aims to integrate image and sound intelligence technology to develop a high-precision, real-time insect recognition system based on CNN, utilizingSVM and LSTM technologies, thereby accurately classifyingimages and sounds and significantly enhancing the efficiency and accuracy of insect monitoring, promoting biodiversity conservation, pest management in agriculture, and facilitating forestry practitioners. The project obtains a rich insect image dataset through theopen-source insect dataset on Kaggle and data support from Dr. Li Yingchao, the director of the Museum of Natural History, combining deep learning algorithms for feature extraction and classification to achieve efficient recognition and monitoring of insect species. The system covers a wide range of insect species, including larvae and adults, and can implement the function of “knowing its name without seeing it” by recognizing the damage caused by insects, enhancing public awareness of insect diversity conservation, enabling precise monitoring and management of agricultural pests, reducing pesticide usage, and protecting the ecological environment.
Team Introduction
This project team consists of students from the 21st and 22nd grades of the School of Information at Beijing Forestry University. Everyone has clear divisions of labor, with awide distribution of personnel and extensive survey areas, working together to advance practical work. The project is divided into four groups: Frontend Group, Data Group, Promotion Group, and Algorithm Group.
Project Progress and Research Plan
Project Progress>>
The Frontend Group has developed two versions of the insect recognition mini-program framework, which include:
Clarify the source of insect information for Interface Three.
Plan One: Call the mini-program cloud database to reference the insect introduction.
Plan Two: Redirect to the corresponding insect information display interface in the mini-program HOME1.
Complete all frameworks except the recognition algorithm interface, and beautify all frontend interfaces as much as possible.
Complete the lazy loading of insect information display and add interesting insect information display features on the campus map of Beijing Forestry University.
The Data Group has found a very comprehensiveInsect Dataset A and organized and listed the existing Beijing Forestry University Insect Collection B in the mini-program cloud server.
Clarify the requirements for images used for model training and collect insect images that intersect between Dataset A and Dataset B.
Select a small number of insect images for model training and observe the results. If the results are good, gradually increase the number of insect images for training.
Resolve the classification order of families (subsequently add families and display them in a specific order); redirect system resources to the Wolai system.
The Promotion Group has completed preliminarysurvey research, created a rich content questionnaire, and disseminated the questionnaire to various professions and age groups through targeted distribution. Currently, the questionnaire is accumulating at a rate of 50 responses per day.
Maintain the public account, divided into two parts: copywriting and layout, and summarize and report weekly work.
After the initial completion of the project, carry out practical deployment and provide assistance for agricultural pest recognition and training of forestry practitioners; find a newspaper to publish the project and let more people recognize and use our mini-program.
Conduct offline practice after project completion, record videos during practice, and lead everyone in patriotic education for team members.
The Algorithm Group has completedthe multi-level display of insects in the mini-program, the insect discussion platform, and all functions of the login interface.
Maintain the public account, divided into two parts: copywriting and layout, and summarize and report weekly work.
Complete the training of all public insect datasets.
Develop a campus map in the mini-program, clicking on areas to display nearby insects, and clicking on insects to show their detailed introduction.
Research Plan>>
Technical Route>>
Project Results
The Frontend Group has achieved significant results in the development of the insect recognition mini-program. The project successfullyconstructed the basic framework and achieved a full process user interaction experience from image upload, waiting for recognition, to result display. When solving the difficulties encountered in image uploading and processing, the Frontend Group ensured the quality and speed of images of different formats and sizes during the upload process by implementing image preprocessing techniques and introducing image compression algorithms. To optimize user experience, the Frontend Groupimplemented loading animations and optimized page layout and interaction design, allowing users to understand the system status in real-time during the recognition process, reducing anxiety.
Additionally, the project hassuccessfully implemented the recognition result display module, including insect names, matching degrees, and detailed introductions, facilitating users’ understanding of recognition results. The addition of a user-upload insect information feature not only enriches the database content but also enhances the interactivity and educational significance of the mini-program. Ultimately, the Frontend Group has created a user-friendly, comprehensive WeChat mini-program, allowing users to take photos or upload insect images for recognition and participate in browsing and adding insect information, contributing to insect research and popularization.
Recognition Interface
Loading Interface
Recognition Result Interface
Swipe left or right to see more
The Data Group successfully completed the preliminary data organization and cleaning work ofInsect Dataset A and Beijing Forestry University Insect Collection B, andclarified the image standards required for model training. Through in-depth communication with the Algorithm Group and Frontend Group, the Data Group fully understood the project’s requirements for insect data, including species, formats, and quality. During the data collection phase, the Data Group overcame challenges such as inconsistent dataset formats and varying data quality, screening and organizing to remove non-compliant data, and standardizing the remaining data toensure data consistency and standardization.
The Data Group maintained close cooperation with the Algorithm Group, continuously adjusting and optimizing the dataset based on feedback, ensuring data validity, and supporting smooth model training. At the same time, the Data Group effectively communicated with the Frontend Group to ensure that the provided data met frontend display requirements, and for data that could not be displayed, adjustments to the display plan or supplementary new datasets were implemented. Through this series of efforts, the Data Group not only provided the Algorithm Group with a high-quality, compliant insect dataset but also enhanced capabilities in information collection, data processing, teamwork, and problem-solving, laying a solid foundation for the success of the entire project.
Swipe left or right to see more
The Promotion Group has played a key role in the project by widely collecting user feedback data through carefully designedsurvey research. This valuable data provides solid support for the subsequent optimization and improvement of the project. Moreover, the person in charge, together with the Promotion Group, regularly holds weekly meetings for project team members on different themes, including “Learning Party History to Enhance Patriotic Ideology Awareness,” “Green Water and Green Mountains are Golden Mountains and Silver Mountains,” and “Safety First in Summer Practice.”
At the same time, the Promotion Group is actively operating the public account, havingsuccessfully published two articles that not only detail the project background, objectives, and significance but also timely share the latest progress and results of the project, effectively enhancing the project’s visibility and user participation. Additionally, the Promotion Group plans to expand the promotion range through various channels, including social media, industry forums, and joint promotions with partners, to ensure the project can reach a wider audience and attract more users to participate in the experience and use the insect recognition mini-program.
Questionnaire
The Algorithm Group has played a crucial role in the project, achieving significant technical breakthroughs and successfullyconstructed an insect recognition system that integrates multi-level display, discussion platform, and login interface, and smoothly completed the training and integration of the model. The in-depth application of image recognition technology, as the core of the project, enables the system to automatically and accurately recognize insect images uploaded by users, significantly improving recognition efficiency and overcoming the time and labor costs of traditional recognition methods.
Utilizing the EasyDL platform based on Convolutional Neural Networks (CNN) for model training, the modeleffectively learned the features of insect images through the effective combination of convolutional layers, pooling layers, and fully connected layers, greatly improving recognition accuracy. During the achievement of project objectives and the resolution of issues, the Algorithm Group faced challenges such as insufficient data volume, the need to improve model accuracy, optimize system response speed, and API calling issues. Throughdataset expansion, model structure optimization, implementation of data augmentation techniques, network request and code optimization, and precise API error handling, the team ensuredthe stable operation and efficient performance of the system.
The project yielded fruitful results, with the Algorithm Group not only training anefficient insect recognition model but also developing ahighly user-friendly mini-program, accumulating rich experience in data processing, model training, and system integration, and enhancing the technical capabilities of the team. Ultimately, the insect recognition system delivered by the Algorithm Group is both practical and highly accurate, possessing significant value in academic research and showing great application prospects in various fields such as agriculture and public health.
Team Building
Each of the four groups has its unique functions and responsibilities, working diligently while actively communicating and learning from each other, collaborating to build ainsect recognition system development team with excellent technology, teamwork, and innovative spirit, together achieving project goals and contributing to biodiversity conservation and sustainable development.
Regular Meetings for Communication and Timely Feedback
Establisheffective communication mechanisms, hold regularteam meetings, and conductcross-group communication meetings to timely exchange project progress and issues encountered, allowing each group to understand the work progress and challenges of other groups, promoting information sharing and mutual understanding, and collaboratively seeking solutions.
Encourage team members to propose innovative ideas and share their expertise and skills, facilitatingtechnology integration and knowledge sharing to enhance the overall professional level of the team and promote the improvement of project technical standards.
Utilize online collaboration tools, such as Kingsoft Documents, and Feishu for file transfers, mutual communication and progress transmission, etc., to enhance the team’s work efficiency and collaboration smoothness.
Screenshot of the First Meeting
Screenshot of the Second Meeting
Screenshot of the Third Meeting
Swipe left or right to see more
Promotion Results
As of August 9, 2023, the public account operated by our project team, “BJFU Insect Recognition System,” has publishedtwo articles, introducing the public to this practical project and releasing weekly summaries, with atotal reading volume of 657 times.
2024 Beijing Forestry University Micro Major Enrollment Guidelines
Information Major Practice | Off to Qigou Forest Farm! Joy and Wisdom Paddle with Us!
END
Planning/Student Union of the School of Information, Learning and Practice Department
Text and Image/ Insect Recognition System Project Team Based on CNN
Reviewers/Fan Shuotian, Jiang Shuai
Editor-in-Chief/Wu Erfu, Zhu Yuning
Editor/Zhuang Jiayu
Review/Wai Hongbin
