摘要
为了通过优化多层感知机权重及偏置提高图像分类决策的正确率,借鉴生物地理学优化(BBO)算法的思想,提出面向图像分类的多层感知机BBO方法.首先提取图像的颜色矩和降维梯度方向直方图(DRHOG)的组合特征作为多层感知机的输入数据,然后使用BBO优化方法得到迭代250次的最优权重及偏置,最后经多层感知机得出测试图像的类别.该方法有效避免算法早熟现象,增强图像分类能力.通过仿真实验,验证此方法比粒子群优化算法优化多层感知机具有更高的求解质量和效率,为图像分类提供又一可行方案.
In order to improve the accuracy of image classification decisions with the MLP＇ s weight and bias optimization, a BBO method for image classification based on multi-layer perceptron is proposed by borrowing from the thought of biogeography based optimi- zation. Firstly, The combination of image features color moment and dimension reduction histogram of oriented gradient is extracted, then put to multi-layer perceptron as input datas. Secondly, BBO method is used to obtain the optimal weights and bias in 250 iterative times. Lastly, the test image ＇ s category is obtained with multi-layer perceptron. This method can avoid algorithm being premature con- vergence and enhance the capability of image classification. With simulation experiments, compared with the method of particle swarm optimization, the higher quality and efficiency of this method is verified. Furthermore, the feasible scheme for image classification is al- so provided .
出处
《四川师范大学学报（自然科学版）》
CAS
北大核心
2015年第6期930-937,共8页
Journal of Sichuan Normal University（Natural Science）
基金
四川省应用基础计划(2013JY0086)
四川省科技创新苗子工程基金(20132033)
关键词
生物地理学优化
多层感知机
颜色矩
降维梯度方向直方图
粒子群优化
biogeography based optimization
multi-layer perceptron
color moments
dimension reduction histogram of oriented gradient
particle swarm optimization
