肖志云,刘洪.小波域马铃薯典型虫害图像特征选择与识别[J].农业机械学报,2017,48(9):24-31.
XIAO Zhiyun,LIU Hong.Features Selection and Recognition of Potato Typical Insect Pest Images in Wavelet Domain[J].Transactions of the Chinese Society for Agricultural Machinery,2017,48(9):24-31.
摘要点击次数: 2534
全文下载次数: 1212
小波域马铃薯典型虫害图像特征选择与识别   [下载全文]
Features Selection and Recognition of Potato Typical Insect Pest Images in Wavelet Domain   [Download Pdf][in English]
投稿时间:2017-05-17  
DOI:10.6041/j.issn.1000-1298.2017.09.003
中文关键词:  马铃薯虫害  小波域  高斯空间模型  特征选择  图像识别  支持向量机
基金项目:国家自然科学基金项目(61661042)和内蒙古自治区自然科学基金项目(2015MS0617)
作者单位
肖志云 内蒙古工业大学 
刘洪 内蒙古工业大学 
中文摘要:为准确、快速地识别马铃薯典型虫害,提出了一种基于小波域的马铃薯典型虫害特征提取与识别方法。该方法以自然环境下的马铃薯虫害分割图像为对象,提取小波域高斯空间模型的高频协方差阵特征值与低频低阶矩(HELM)的12个不变纹理特征、空间域Hu不变矩的4个形状特征,进行支持向量机(SVM)的虫害分类识别。通过对8类典型虫害的识别,试验结果表明:在SVM识别方法下,本文HELM特征提取方法,相比传统纹理特征提取方法,在特征计算量不增加的同时,平均识别率至少提高了17个百分点;在HELM特征与Hu矩特征下,本文SVM的运行时间为0.481s,比人工神经网络快了近2s,平均识别率为97.5%,比人工神经网络、贝叶斯分类器识别率提高了至少6个百分点,有明显的识别优势。
XIAO Zhiyun  LIU Hong
Inner Mongolia University of Technology and Inner Mongolia University of Technology
Key Words:potato insect pests  wavelet domain  Gaussian space model  feature selection  image recognition  support vector machine(SVM)
Abstract:In order to recognize potato typical insect pests accurately and quickly, a new feature extraction and recognition method based on wavelet and space domain was proposed. The processing object in the method was the segmented image of insect pests separated from complex background by the two dimensional Otsu method and morphological method. Aiming at the processing object, totally 12 invariant texture features of high frequency covariance matrix eigenvalues and low frequency lower order moments (HELM) were extracted from the high frequency images in the horizontal, vertical and diagonal directions, forming a Gaussian space model, and from low frequency image decomposed by sym8 wavelet function. Meanwhile, 4 Hu moments with invariant shape features were extracted from the binary image of the processing object. As thus, 16 pest features were put into support vector machine (SVM), and the results of insect pest classification could be obtained. For SVM classifier, the One-vs-One voting strategy was adopted, and the parameters, including radial basis kernel function parameter, error cost coefficient and relaxation coefficient were set to 0.0125, 60 and 0.001, respectively. By the classification of 8 kinds of pests, on the one hand, using the same SVM method, the test results showed the effectiveness of proposed HELM feature extraction. Texture features in wavelet domain were traditionally related to single scale low frequency lower order moments (SLM), including the mean, variance and the third order moment of low frequency image, multiscale low frequency lower order moments (MLM), multiscale high frequency lower order moments and low frequency lower order moments (HMLM), and LBP features for the low frequency image. Texture features in space domain were traditionally related to LBP, PCA and features based on gray-level co-occurrence matrix (GLCM). Compared with SVM recognition rates of the traditional texture features in wavelet domain and space domain, it was found that the proposed HELM feature had a higher recognition rate which were increased by at least 17 percentage points. In addition, the proposed HELM feature had moderate run time of 11.7 s containing from features extraction of 210 pest images to SVM classification of 8 kinds of typical pests. On the other hand, using the same HELM features and Hu moments, the test results showed the effectiveness of the proposed SVM recognition. For artificial neural network (ANN), three layers BP network structure was constructed and the sigmoid transfer function of hidden layer was selected. For Bayes classifier, Gaussian window function was used for estimating probability density. Compared with ANN run time, containing from the train for 105 pest images to the test for 105 pest images, the run time of the proposed SVM was 0.481s, nearly 2s less than ANN. Meanwhile, compared with ANN and Bayes recognition rates, the proposed SVM recognition rate was 97.5% , increasing at least 6 percentage points.

Transactions of the Chinese Society for Agriculture Machinery (CSAM), in charged of China Association for Science and Technology (CAST), sponsored by CSAM and Chinese Academy of Agricultural Mechanization Science(CAAMS), started publication in 1957. It is the earliest interdisciplinary journal in Chinese which combines agricultural and engineering. It always closely grasps the development direction of agriculture engineering disciplines and the published papers represent the highest academic level of agriculture engineering in China. Currently, nearly 8,000 papers have been already published. There are around 3,000 papers contributed to the journal each year, but only around 600 of them will be accepted. Transactions of CSAM focuses on a wide range of agricultural machinery, irrigation, electronics, robotics, agro-products engineering, biological energy, agricultural structures and environment and more. Subjects in Transactions of the CSAM have been embodied by many internationally well-known index systems, such as: EI Compendex, CA, CSA, etc.

   下载PDF阅读器