Flower Maintenance System Based On Embedded System And Deep Learning

With the development of Internet of Things and deep learning technologies,more and more appliances are developing in the direction of intelligence and humanization,and more and more people are growing flowers as their living standards and life interests improve.To solve people’s problems in flower maintenance,a maintenance system is developed combining automatic irrigation and flower health recognition by combining technologies of Io T,embedded system and deep learning.It is also proposeds a lightweight flower recognition algorithm based on knowledge distillation and MobileNetV3 in this thesis,so as to solve the problems that traditional flower recognition algorithms are not suitable for deployment in mobile device and high cost of flower dataset annotation.The main work and innovations are as follows:(1)To solve the problem of inconvenient watering in flower maintenance,the automatic irrigation system is designed,which can realize the functions of automatic system networking,mobile control,multichannel parallel,soil temperature and humidity sensing,real-time watering,timing watering and low water automatic watering by relying on Ali cloud IOT platform.(2)To solve the problem that it is difficult to judge the health condition of flowers in flower maintenance,MobileNetV3 is used as a platform to build a lightweight flower recognition model,which can be deployed with mobile terminal and has the functions of identifying flower species and judging the health condition.(3)Integrating automatic irrigation,information sensing,remote control and health recognition functions on mobile through application software,thus realizing a complete flower maintenance system.(4)A exponential decay knowledge distillation(EDKD)algorithm is proposed whitch is higher effectiveness than conventional knowledge distillation algorithm,EDKD is applied in the system to the training of flower recognition model to improve the accuracy of the model.(5)To address the problem of high cost of flower data collection and labeling in actual industrial production,a label-free flower dataset is collected and constructed,and a flower recognition method based on no label knowledge distillation(NLKD)is proposed,which can use a large amount of easily accessible label-free data to assist the training of flower recognition models to improve the accuracy of the models or save labor costs.The physical verification of the system and algorithm simulation results show that the automatic irrigation system designed in this thesis can effectively solve the problem of inconvenient watering in flower maintenance,and the flower identification algorithm proposed in this thesis has more advantages in terms of performance,efficiency and cost saving compared with traditional methods.