Research on Fish Intelligence for Fish Trajectory Prediction Based on Neural Network

This paper researches the behavior modes of some intelligent creature in some environment. The gained modes are used as movement models to construct NN to predict the moving trajectory and then catch it. Firstly the behavior patterns of fish that kept trying to escape from the net attached at robot's hand were studied through lots of experiments. The patterns were divided into five sorts and the learning procedures were divided into three stages. Based on this, the position, orientation and speed of each time were used as the input of multi layer perceptron (MLP) neural networks (NN), and the positions of the fish at next time were the outputs. The NN adopted extended delta-bar-delta (DBD) algorithm as learning method. Thus the NNs were constructed to study the moving regulations of fish in every pattern to predict the moving trajectory. The simulation results shows that the BP NN constructed here have the advantage of faster learning rate, higher identifying precision and can predict the fish trajectory successfully. The research is significant for visual servo in robotic system.