四元数和脉冲耦合神经网络应用于足球检测

提出一种基于四元数傅里叶变换注意力选择和脉冲耦合神经网络在图像中跟踪足球的方法. 首先进行图像预处理以去除球场以外的区域，用四元数注意力选择算法提取感兴趣区域，基于颜色、形状、面积等多种特征检测足球. 若检测失败，则采用卡尔曼滤波器预测足球位置. 仿真结果表明，与基于速度控制的动态卡尔曼滤波和实时足球检测两种方法相比，检测成功率分别提高9.6%和14.9%.

Soccer Detection in Images Based on Quaternion and Pulse Coupled Neural Network

This paper proposes a soccer detection method that combines the attention selection model of phase spectrum of quaternion Fourier transform (PQFT) and pulse coupled neural network (PCNN). In the preprocessing, the region outside the field is removed, and the region of interest extracted using PQFT. The
target is detected according to the physical characteristics such as color, shape and size. If no candidate or more than one are detected, a Kalman filter is used to make prediction. Simulation shows that the identification rate is improved by 9.6% and 14.9% respectively as compared to the dynamic Kalman filtering with velocity control and the real time ball detection framework introduced in the literature.