Challenges

In the Camellia oleifera production areas of southern China, Thailand and Vietnam, the damage caused by pests and diseases has become increasingly severe recently. Traditional pest and disease identification relies on human experience, which is inefficient and prone to misjudgment, resulting in lagging prevention and control, yield loss and pesticide abuse, threatening ecological security and farmers’ livelihoods. This project builds an intelligent recognition system via Artificial Intelligence (AI) technology to achieve real-time and precise diagnosis of pests and diseases, facilitating early intervention and green prevention and control. It promote digital transformation and innovation in agroforestry through AI (SDG 9). It reduce overuse of pesticides to promote sustainable agricultural practices (SDG 2). It directly supports the agricultural efficiency enhancement goal of SDG and indirectly promotes the ecological balance of SDG 15, protecting biodiversity by reducing the use of chemical pesticides. Technology promotion can also enhance the resilience of smallholder farmers (SDG 1) and promote sustainable agricultural practices.

Toward a Solution

The project aims to utilize AI-driven pests and disease identification technology for Camellia oleifera to solve the problems of lagging prevention and control, yield loss, and pesticide misuse in the Camellia oleifera producing areas in the southern region of China, Thailand, Vietnam, and other northern Camellia oleifera producing areas, which are caused by the difficulty of identifying pests and disease for Camellia oleifera, and the inefficiency of the traditional manual identification and the high misclassification rate. Enhance technology adaptation for smallholder farmers to stabilize livelihoods and reduce the risk of disaster-induced poverty (SDG 1); Increase Camellia oleifera yields and safeguard food security and cash crop revenues by reducing pest losses (SDG 2); Promote digital transformation and innovation in agroforestry through AI (SDG 9); Reduce overuse of pesticides to promote sustainable agricultural practices (SDG 12); Reduce the damage caused by chemical pesticides to soils and ecosystem damage by chemical pesticides and protect biodiversity (SDG 15).

By establishing a list of Camellia oleifera pests and diseases in the Lancang-Mekong region, we have constructed an identification model that combines semi-supervised automatic data labeling methods and the mechanism of “user feedback-expert review-iterative optimization”, and supports multilingual interfaces (Chinese, Vietnamese and Thai), thus reducing misjudgments due to the differences in regional nomenclature, lowering the cost of manual labeling, and breaking the language barrier. To promote the sharing of knowledge and experience in prevention and control among farmers and agricultural technicians in China, Thailand, and Vietnam, user feedback and expert review are combined to continuously iterate the model to enhance the long-term applicability of the system, and to jointly accelerate and realize SDG 1, SDG 2, SDG 9, and SDG 15.

The partners from China, Vietnam and Thailand have realized in-depth synergy through multi-level interaction, and researchers from Vietnam Forestry University and Queen Sirikit Botanical Garden of Thailand have participated in the case implementation and jointly investigated pests and diseases of both Camellia oleifera and tea tree in Vietnam and Thailand. China, Thailand and Vietnam have a clear division of labor among scientific and technical personnel, Guangxi Forestry Research Institute (GFRI) is responsible for the survey of Camellia oleifera pests and diseases in China, data collection, research and construction of intelligent identification models and identification software development, experts from Vietnam Forestry University and GFRI survey Camellia oleifera pests and diseases in Vietnam in provinces such as Lang Son, Nghe An and Quang Tri, and the Queen Sirikit Botanical Garden of Thailand, Thailand’s Kunkchat Thanar Foundation, and experts and technical staff of Queen Sirikit Botanical Garden of Thailand and Kunchai Thana Foundation of Thailand and experts of GFRI jointly surveyed Thai pests and diseases of both Camellia oleifera and tea tree and in Chiang Rai, Chiang Mai and Nan provinces of Thailand.

Some results have been achieved in SDG. Forestry scientists from China, Thailand and Vietnam have jointly conducted pests and diseases surveys in Camellia oleifera producing areas in southern China, Chiang Rai, Chiang Mai and Nan provinces in Thailand, and Liangshan, Beijiang and Nghe An provinces in Vietnam, identifying 230 species of Camellia oleifera pests and diseases (SDG 15). GFRI combined with AI to develop the “Lancang-Mekong Region Camellia oleifera pests and disease Intelligent Identification APP”, which can identify 82 pests and diseases species, and the identification has changed from expert identification to intelligent identification, shortening the identification time to 3 seconds (SDG 9). In the China-Thai Camellia oleifera cooperation base in the “Golden Triangle Region” of Mae Fah Luang District, Chiang Mai Province, Thailand, through on-site technical training, the base technicians and foresters have gone from being at a loss for Camellia oleifera pests and disease to being able to identify pests and diseases and obtain prevention and control suggestions by taking a picture with their smartphones, which has resulted in timely control of disasters and reduced the loss of Camellia oleifera production by more than 15%, significantly improving the economic benefits (SDG 1 and SDG 2). At the base of Guangxi Yiyuan Camellia oleifera Industry Development Co., Ltd, the management personnel identified pests and diseases in the breeding base of scented flower Camellia oleifera seedlings and plantations and took measures in time through APP, which reduced the incidence of pests and diseases by 21% and improved the quality of Camellia oleifera (SDG 1 and SDG 2).

Improving identification algorithms via self-built YOLOv7 and “user feedback-expert review-iterative optimization” mechanism for case innovation. The establishment of universal Camellia oleifera pest and diseases intelligent identification software (local language version) is promoted in the form of public welfare, which can quickly and intelligently identify Camellia oleifera pests and diseases, and give prevention and control suggestions to reduce disaster losses in the three countries and improve the country’s competitive advantage in this area. The successful implementation of this case has brought a good demonstration effect, promoting the cooperation between China and ASEAN Lancang region forestry artificial intelligence to go deeper and more practical: (1) To facilitate the signing of “the Memorandum of Understanding between Guangxi Forestry Research Institute and Vietnam National Forestry University on Cooperation on Artificial Intelligence Enabling Forestry Development on February 19, 2025; (2) Facilitating the establishment of three projects involving artificial intelligence in forestry, such as the “Intelligent Identification of Pests in Eucalyptus and Camellia oleifera in the Lancang-Mekong Region”, a project of the Platform Center of the Ministry of Science and Technology in 2023, and expanding the cooperation between China and ASEAN countries on artificial intelligence to include the identification of eucalyptus, anise, incense and other pests and diseases.

The good practices of this project are highly replicable and adaptable. First, the technology system adopts a modular design, and the identification algorithms can be adapted to the identification of pests and diseases of different cash crops such as eucalyptus, tea tree, coffee, rubber, etc. Second, the governance mechanism is resilient, and the cross-border collaboration framework can be adapted and applied to the identification of eucalyptus pests and diseases in China-Vietnam and Laos, and can be put into operation with only the replacement of 40% of the localized data. Three key conditions need to be met to ensure replicability: (1) Infrastructure (at least 3G network coverage); (2) More than 20,000 labeled sample data at the initial stage; (3) A joint prevention and coordination mechanism involving multiple parties. The key points for adaptation are: the establishment of a 60-day rapid iteration mechanism led by local experts (i.e., 2 months to complete local data collection and model fine-tuning), and the development of a configurable multi-language interface.