Author：Lin Yu-Heng/Assistant Researcher, Taichung District Agricultural Research and Extension Station
Chen Wei-Ling/Associate Research Fellow, Taichung District Agricultural Research and Extension Station
Shen Chun-Jung/ Research Assistant, Taichung District Agricultural Research and Extension Station

Broccoli (Brassica oleracea L. var. italica) is an important vegetable crop worldwide. In Taiwan, it is mainly produced under contract farming systems and is consumed both fresh and as frozen processed products. The major broccoli cultivation areas in Taiwan include Changhua, Yunlin, Chiayi, and Kaohsiung. Thanks to recent initiatives of Green Environmental Payment Program and the promotion of contract farming for processing broccoli along with standardized frozen processing and storage operations by the Agriculture and Food Agency, the cultivation area of broccoli has shown a steady annual increase. The industry cluster has formed through the integration of contract farming, collective marketing (including cut-vegetable supply for institutional meals), and processing plants (including farmer groups) with inventory adjustment capabilities.

To address the longstanding challenges in production technology and industry development, this study integrates growth monitoring tools, builds a management platform for predicting broccoli harvest period and yield, and develops an AI-assisted harvesting machine for broccoli heads. The aim is to achieve labor-saving cultivation and support risk assessment, management decision-making, and production-marketing planning for broccoli.

This study collected continuous data over 24 planting cycles from 2020 to 2022, including broccoli plant growth, photosynthesis-related physiological parameters, UAV-based spectral vegetation indices, AIoT environmental data, and final yield. After data cleaning, integration, and transformation, machine learning was conducted using a convolutional neural network (CNN) framework. The trained model was validated with over 190 farm records covering approximately 53 hectares from Changhua, Yunlin, and Chiayi counties during 2022–2023 under TGAP certification, accumulating more than 1.23 million data points for analysis. Results showed that the predicted flowering and harvest dates deviated by 0 and ±4 days respectively, with a yield prediction accuracy of 94%. By planting at the optimal predicted times, yields increased by 20.3% to 32.1% compared to the average yield in the same period. Additionally, the UAV-NDRE index measured before harvest exhibited a strong linear correlation with final yield (R2 = 0.743). With ongoing optimization of the dynamic broccoli growth prediction model, an integrated data management and collaborative SaaS platform has been developed to facilitate stable production planning and supply chain management for broccoli.