深度果蝇神经网络及其实时能见度预测

Deep fly neural network and its real-time visibility prediction

能见度预测属于时间序列预测领域的问题,如何实现对时序数据的处理以及捕获数据中的时间长期依赖关系,是当前研究人员的主要关注焦点。针对能见度预测模型构建难的问题,依据果蝇视觉系统在图像感知、学习和信息反馈过程中的生物学原理,提出一种基于视频监控图像的深度果蝇神经网络。首先,通过设计前馈果蝇视觉神经网络,提取图像的能见度特征,进而将特征信息送入多层感知器,获得关于未知权值和阈值的能见度与视频帧的映射关系;其次,借助梯度下降法获得在线预测能见度的深度神经网络。实验结果表明,该神经网络能实时、有效、准确地预测雾天环境下的能见度,预测效果稳定且精度高,具有较好的应用潜力。

Visibility prediction is a problem in the field of time series prediction,and how to implement the processing of time-series data and capture the temporal long-term dependence in the data is the main focus of current researchers.In order to solve the difficulty of constructing visibility prediction models,a deep fly neural network,based on video surveillance images is proposed in terms the biological principles of the fly visual system such as image perception,learning and information feedback.In the design of the model,a feedforward fly visual neural network is designed to extract the image-based visibility feature,and then the acquired feature information is fed into a multi-layer perceptron to obtain the mapping relationship between visibility and video frames with regard to unknown weights and thresholds.These,together with the gradient descent method,derive the deep neural network for real-time visibility prediction.The comparative experimental results show that the neural network can accurately predict the visibility of the foggy environment,and has great application potential.