INTELLIGENT FRAMEWORK FOR MONITORING CROPS IN GREENHOUSES
Wireless Sensor Networks (WSNs) and Wireless Visual Sensor Networks (WVSNs) are two monitoring technologies that have the potential for use in many application domains, and both are poised for growth in many markets from the farm to the office. Integrating a WSN with a WVSN in a commercial greenhouse setting is an application domain yet to be researched. The integration of WSN and WVSN has the potential to overcome the problems other monitoring systems have encountered. A system combining these two wireless networks will need no human interaction, deliver real-time data indicating an adverse event, be cost-effective, and use less power. Additional efficiencies specific to the greenhouse application are needed due to its clutter and occluded environment, very large area, and restricted energy plan. This thesis presents a framework that combines WSN, WVSN, Machine Learning (ML), deep learning, and image processing to address the challenges faced by operators of commercial greenhouses. The framework achieves three objectives. First, finding the optimal placement of WVSN nodes to minimize the number of installed camera nodes. Second, monitoring the growth of the plants and detect any abnormalities caused by pests or diseases. Third, controlling the microclimate inside the greenhouse and dynamically predicting the duty cycle activities of the monitoring sensors. Our first objective is achieved by formulating and solving an optimization problem to find the best placement for the camera sensors of the WVSN, maximize the area covered, and minimize the number of camera sensors used with good quality images. The second objective is achieved by using the Hough Forest ML and image processing techniques on the images taken by the WVSN to detect any fungus, monitor the growth of the plant, and to increase crop production and quality. Our third objective is achieved by controlling the microclimate inside the greenhouse using deep learning prediction Long Short-Term Memory (LSTM) model. The prediction model will not only control the microclimate inside the greenhouse but also predict and control the monitoring sensor’s duty cycle to decrease energy consumption and prolong the network’s lifetime.