Vision-Based Autonomous Greenhouse

An autonomous controlled-environment system built around an ESP32 sensor/actuator network and a TensorFlow Lite vision model. Cameras assess plant health and growth stage while environmental sensors track temperature, humidity, and light; a control loop drives irrigation valves, grow lights, and ventilation over MQTT to keep the greenhouse within crop-specific setpoints.

Manual greenhouse management is reactive and labour-intensive, and simple thermostat-style controllers ignore the actual state of the plants. The goal is a controller that sees plant condition and regulates the environment proactively.

An ESP32 network gathers temperature, humidity, and light readings while a TensorFlow Lite vision model assesses plant health and growth stage from camera frames. A control loop fuses both into actuator commands — irrigation, lighting, ventilation — published over MQTT, holding crop-specific setpoints automatically.

Planned — targets: stable maintenance of environmental setpoints, reduced manual intervention, and measurable water/energy efficiency versus rule-of-thumb control.

1. Phase 1

   Sensing & actuation

   Build the ESP32 sensor/actuator network and MQTT control bus.

2. Phase 2

   Vision model

   Train a TensorFlow Lite model for plant-health and growth-stage assessment.

3. Phase 3

   Control loop

   Fuse sensor and vision inputs into setpoint-tracking actuator control.

4. Phase 4

   Tuning

   Optimise for crop health, water, and energy use across grow cycles.