A system for plant detection using sensor fusion approach based on machine learning model

There is a need for new precision agriculture approaches that allow real-time intervention within sugarcane rows to reduce costs and minimize negative environmental impacts. Therefore, our goal was to test an alternative system for detection within rows of sugarcane plants. The objective was to determine the errors of an alternative system to detect targets of different sizes at different travel speeds. A system with a photoelectric sensor, ultrasonic sensor, and encoder was developed to detect and map the plants within the sugarcane row. The use of sensors separately and simultaneously for plant detection was compared. To improve the accuracy of plant detection, decision tree (DT), random forest (RF), and support vector machine (SVM) models were tested. The three machine learning (ML) models used data generated by the photoelectric and ultrasonic sensors along with the displacement sensor. The models were compared in terms of their precision in detecting plants within sugarcane rows. The approach with the two sensors and the DT model had the best precision (>90%) in plant detection. The sensors have the ability to detect 91% of the total plants (recall = 0.91). The travel speed influenced the performance of the sensors in detecting the targets, especially at 2 m s(-1). The accuracy and precision of the plant detection system at different speeds indicate the possibility of the sensors being used in integrating systems for real-time interventions. In addition, the system generates data to plant and gap maps, which are useful information to support the site-specific management of sugarcane fields.

Automated summary

The study aimed to test an alternative system for detecting sugarcane plants within rows and compare the accuracy of different machine learning models. By using a combination of photoelectric sensor, ultrasonic sensor, and encoder, the research found that the approach with two sensors and the decision tree model had the best precision in plant detection.