When crops meet machine vision: A review and development framework for a low-cost nondestructive online monitoring technology in agricultural production

Abstract

The Food and Agriculture Organization (FAO) has indicated that digital technology is key for improving the resilience of food systems. Smart models have been developed for agricultural water, fertilizer, medicine, and environmental regulations, in which data-driven quantity and precision are crucial. However, data acquisition methods based on manual observation cannot meet the requirements of large amount of real-time data. The development of machine vision provides a new method for online non-destructive monitoring. We discuss algorithm types and evaluation methods for machine vision applications based on RGB images considering their low cost and easy access. This paper reviews progress in the application field, covering the entire process from planting to postharvest, and the application of sensing and control equipment in agricultural practice. Finally, aiming at the problems such as lack of agricultural data set, poor model portability, and large model size, a new algorithm framework based on “data layer - model layer - deployment layer,” multi-parameter “environmental data - image data” and multi-method fusion of “mechanism model - machine vision” was proposed to provide a basis for low-cost nondestructive online crop monitoring.