Visual navigation control for agricultural robot using serial BP neural network

A visual navigation system for agricultural robot was developed to navigate the robot moving autonomously through field visions. To do this, it was necessary to use those sufficient and effective navigation parameters. Previews and feedback parameters were used in a visual control system to navigate by agricultural robot. The current ROI window was divided into an upper and lower path region to obtain the previewing and current navigation information. Based on the previewing, current and previous cycle navigation parameters, a serial BP neural network was trained to adjust the link weight coefficients and the threshold of each neuron, to ensure a perfect output of navigation parameters. The visual control system was validated using a serial BP neural network and satisfactory steering control results were obtained. Maximum feedback deviation of abscissa position between the actual and ideal target path was –0.069 m and the maximum previewing deviation of abscissa position was –0.043 m. Maximum angular feedback deviation was –3.5°, and maximum angular previewing deviation was –2°. Experimental results showed that the proposed method could obtain high accuracy navigation parameters.