Development and testing of a ground recognition system for tractor field operations

Abstract

To overcome the problems of poor recognition accuracy, slow response speed, and unsuitable for farmland ground environment based on the traditional ground recognition method of vehicle mathematical model, this paper proposes a ground recognition system for tractor field operation based on wheel acceleration. A wheel acceleration sensor and data acquisition device are designed, and a ground recognition algorithm based on Principal Component Analysis - Genetic Algorithm optimization of an Extreme Learning Machine (PCA-GA-ELM) is investigated. The system can collect acceleration samples and extract statistical features in different frequency bands in real time according to the cycle of wheel rotation and then use the trained recognition model to recognize the ground type. In the actual operation process, this system only needs to collect the acceleration test data of one rotation of the tractor wheels to achieve the recognition of the ground. The test set data show that this system achieves 96 % discrimination accuracy, higher than the 87 % of the ELM model without genetic algorithm optimization and 92 % of the GA-BP neural network model. In addition, the real-time ground recognition test shows that the increase in operation speed will slightly reduce the accuracy of the recognition model, but the overall can still maintain 94–98 % accuracy, indicating that the system has good stability. This system can provide a reference for real-time speed adjustment, implement position, and other control parameters during tractor operation.