Throughout the day as the sun evaporates the water from a plant’s leaves via a process called transpiration, observers will notice that they tend to get a little bit droopy. Also known as drought stress, this response to a loss of water results in low turgidity (internal water pressure) and can impact the ability of the plant to grow correctly. Traditional irrigation monitors use soil moisture sensors to determine the soil’s water levels, but Terry Rodriquez and Salma Mayorquin wanted to create something a bit more unique: a visual droop detection system. 

Their device, which they affectionately call the “Droop, There It Is,” features a Nano 33 BLE Sense and ArduCam camera module to take pictures of the plant and uses an image classifier to determine if the plant is drooping or not. They started by taking a pre-trained MobileNetV2 base model and fine-tuned it with a set of 6,000 images. After optimizing the result with grayscale reductions and knowledge distillation techniques, the team deployed it onto their Nano 33 BLE Sense for inferencing. 

Although the device only signals when the plant needs water over Bluetooth® Low Energy for now, it can be augmented in the future to directly control pumps and valves if needed. This project is a great demonstration of how machine learning can be harnessed to reduce overwatering and increase efficiency. You can read more about it here or check out their video below!