基于全卷积神经网络的计算机视觉目标检测算法

Computer vision target detection algorithm based on full convolution neural network

针对传统计算机视觉目标检测算法准确率较低的问题,提出基于全卷积神经网络的计算机视觉目标检测算法.通过采样、量化以及编码将计算机视觉图像转换为数字图像,经图像平滑、图像增强、目标图像截取一系列处理后完成目标图像特征抽取,实现图像预处理;利用全卷积神经网络训练完成计算机视觉图像目标检测.结果表明,采用所提算法检测计算机动态图像及静态图像时,准确率和召回率分数均更接近1,证明算法的检测准确率更高,且检测耗时较少.

Aiming at the problem of low accuracy of traditional computer vision target detection algorithms,a computer vision target detection algorithm based on full convolution neural network was proposed.The computer vision images were converted into digital images through sampling,quantization and encoding.After a series of processing of images,e.g.,smoothing,image enhancement and target image interception,the feature extraction of target images was completed,and the image preprocessing was accomplished.The target detection of computer vision images was completed by full convolution neural network training.The results show that the accuracy and the recall score are both closer to 1 when using the as-proposed algorithm for the detection of computer dynamic and static images,confirming that this algorithm has higher detection accuracy and shorter detection time.