Model Predictive Control Of Unmanned Aerial Vehicle For Locust Detection And Bio-pesticide Spraying

Swarm of Locust are very harmful for food security, quality and quantity of agriculture products.rnEthiopia is one of the countries which is extensively affected by locust invasion. The locustrnswarms have destroyed large swaths of food and pasture in Ethiopia which lead to famine and displacingrnthousands of people from their home. Ethiopia battled the swarms by spraying pesticidesrnfrom air using helicopters leased from FAO. With this consideration, precise locust detection andrnbio-pesticide spraying is significant for preventing locust plagues.rnThis thesis is going to focus on the design of Model Predictive Control of UAV for locust detectionrnand bio-pesticide Spraying. To accomplish this design: First the dynamics of the system wasrnunderstood then the mathematical model of the system was done and it was based on an agriculturernspray drone (JMR-X1400). The Newton-Euler formalism was used to model the dynamic systemrnand verified in Simulink. The flight controller is designed and MPC is implemented for this thesis.rnFor this non-linear dynamic system of a quad-copter NMPC (non-linear MPC) is chosen. Multiplernshooting method is selected to transform the optimal control problem to nonlinear program (NLP).rnTo solve the NLP, CasADi in MATLAB is used and the solver is Ipopt (Interior Point Optimizer).rnThe NMPC was able to control the quad-copter, which means the quad-copter was able to followrnthe given reference trajectory with minimum control effort. Since the quad-copter is used to sprayrnpesticide, there will be a change in mass when it sprays. For this reason the Recursive Least SquarernEstimation (RLSE) is used to estimate the mass change and the model can be updated using thernestimation. The proposed method works adequately. The RLSE was able to estimate the massrnchange and the quad-copter was still able to track the reference.rnManual monitoring is a labor-intensive job and expensive for large farms. To tackle this problem,rnimage recognition have provided a promising solution for detecting pests. So for this thesisrnImage recognition system is developed to detect and recognize the Locust swarm. Since it is anrnImage classification, CNN is chosen and the programming language is in python. After passingrnthrough different procedures the final training accuracy of the machine is 95:19%.

Get Full Work

Report copyright infringement or plagiarism