Real-Time Strawberry Plant Classification and Efficiency Increase with Hybrid System Deep Learning: Microcontroller and Mobile Application

Abstract

The strawberry plant has three life stages: seedling, blooming, and crop. It needs different acclimatization conditions in these life stages. A dataset consisting of 10,000 photographs of the strawberry plant was prepared. Using this dataset, classification in convolutional neural networks was performed in Matrix Laboratory (MATLAB). Nine different algorithms were used in this process. They were realized in ResNet101 architecture, and the highest accuracy rate was 99.8%. A low-resolution camera was used while growing strawberry plants in the application greenhouse. Every day at 10:00, a picture of the strawberry plant was taken. The captured image was processed in ResNet101 architecture. The result of the detection process appeared on the computer screen and was sent to the microcontroller via a USB connection. The microcontroller adjusted air-conditioning in the greenhouse according to the state of the strawberry plant. For this, it decided based on the data received from the temperature, humidity, wind direction, and wind speed sensors outside the greenhouse and the temperature, humidity, and soil moisture sensors inside the greenhouse. In addition, all data from the sensors and the life stage of the plant were displayed with a mobile application. The mobile application also provided the possibility for manual control. In the study, the greenhouse was divided into two. Strawberries were grown with the hybrid system on one side of the greenhouse and a normal system on the other side of the greenhouse. This study achieved 9.75% more crop, had a 4.75% earlier crop yield, and required 8.59% less irrigation in strawberry plants grown using the hybrid system.

Keywords

• deep learning
• convolutional neural networks
• MATLAB
• hybrid system
• mobile application
• productivity