The purpose is to mitigate network congestion (NC) and high energy consumption (EC) in the traditional Internet of Things (IoT)-supported crop monitoring system (CMS). Firstly, the current work summarizes the status quo of IoT and edge computing (EC) technologies. Secondly, it constructs an intelligent multi-sensor-based real-time CMS. Consequently, an EC-based agricultural IoT (AIoT) architecture is proposed. Finally, the current work optimizes the task scheduling at the IoT edges using deep reinforcement learning (DRL) and proposes the DRL-optimized EC-AIoT-based CMS. Furthermore, the performance of the proposed DRL-optimized EC-AIoT-based CMS is verified through experiments. The results show that: (1) There is little difference between the data collected by the proposed CMS and the manual measurement, so the proposed CMS has a high data accuracy. (2) The performance of the DRL-optimized real-time scheduling model is better than the traditional methods in both scheduling time and data integrity. (3) Under the proposed EC-AIoT-based CMS, the server occupancy and queueing time are significantly lower than other algorithms. The purpose is to provide important technical support (TS) for improving the efficiency and quality of crop monitoring and agricultural green productivity (GP).