基于BF神经网络的图像智能识别方法  

INTELLIGENT IMAGE RECOGNITION METHOD BASED ON BF NEURAL NETWORK

作　　者：崔玲玲[1] 张家明 代江艳 Cui Lingling;Zhang Jiaming;Dai Jiangyan(School of Computer Engineering,Weifang University,261061,Weifang,Shandong,China;Jihongtan Reservoir Management Station of Shandong Water Diversion Project Operation and Maintenance Center,266100,Qingdao,Shandong,China)

机构地区：[1]潍坊学院计算机工程学院,山东潍坊261061 [2]山东省调水工程运行维护中心棘洪滩水库管理站,山东青岛266100

出　　处：《山东师范大学学报（自然科学版）》2023年第3期270-277,共8页Journal of Shandong Normal University(Natural Science)

基　　金：国家自然科学基金资助项目(62006174)。

摘　　要：目前的图像智能识别方法中,由于干扰因素过多,导致图像识别的结果误差值较大.为此提出基于BF神经网络的图像智能识别方法。采用传感器装置采集静态图像样本和动态图像样本,经过拜尔滤光片后,导入到图像采集卡中,由图像采集卡将图像分解成为RGB模式下的三种原色的原始图像。采取HOG特征算法对图像样本的特征进行预处理,分别完成正向传播和后向传播,灰度化图像样本的检测窗口,通过计算像素梯度,进一步排除弱化光照等因素的干扰,计算重叠区域的权重投影,并归一化对比度,最终得出可供识别的特征向量。基于BF神经网络构建图像智能识别模型,采用梯度下降法对模型的输出层和隐藏层进行权值调整,采用正负样本1∶3的比例对该模型进行训练。设计对比实验检测该方法的可行性,将实验结果与BP神经网络识别方法和线性神经网络识别方法进行对比,得出该方法在不同识别次数下,识别误差值均小于其他两种方法的结果,能够有效提高图像智能识别的精准度。At present,most existing image intelligent recognition methods exhibits large error because of too many distraction.Therefore,an intelligent image recognition method based on BF neural network is proposed.First,the static and dynamic image samples are collected by the sensor device,and then imported into an image acquisition card after passing through Bayer filters,which can decompose the image into three original colors in RGB mode.Second,the HOG feature is extracted to preprocess the image samples,and the forward propagation and backward propagation are respectively completed.Particularly,in the detection window of gray-scale image samples,the pixel gradient is calculated to further eliminate the interference of factors such as weakening light.The weight projection of the overlapping region is calculated,and the contrast is normalized,and finally the recognizable feature vector is obtained.The intelligent image recognition model based on BF neural network is constructed,in which the weight of the output layer and the hidden layer is adjusted by the gradient descent,and the ratio of positive and negative samples is 1:3 for training.Comparative experiments are designed to test the feasibility of the proposed method.By comparing the proposed method with the BP neural network recognition method and the linear neural network recognition,it indicates that the recognition error of the proposed method is smaller than the results of the other two methods with different recognition times,which can effectively improve the accuracy of intelligent image recognition.

关 键 词：图像识别 智能识别 BF神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]