基于图像识别的储粮害虫分类的研究
|
摘要
本文系河南省自然科学基金和中科院模式识别国家重点实验室开放课题资助项目。
在粮食储藏期间，因害虫危害所造成的损失十分严重，怎样有效防治储粮害虫，一直都是世界各国粮食储藏行业的研究热点。早在1972年，联合国粮农组织就提出了建立在生态学基础上的害虫综合防治策略(Integrated Pests Management)，而害虫的检测是IPM体系的重要组成部分。只有准确的检测，才能做到有目的的防治，既不会因虫害造成损失，也不会因盲目防治造成浪费，加重对粮食、环境的污染。
目前，我国主要采用人工扦样的方法来检测储粮害虫，由于扦样法技术落后、工作量大、效率低、还受工作人员主观因素的影响，因此难以适应我国粮食储藏工作的需要。为适应我国储粮现代化建设的需要，我们提出了基于图像识别的在线检测储粮害虫的新思路，围绕该课题本文主要进行了以下几个方面的研究工作：
1．图像识别储粮害虫检测系统的软硬件设计。分析比较了现有几种主要的储粮害虫检测方法，指出进行储粮害虫图像识别在线检测研究的意义；给出了检测系统的具体方案，软件和硬件的设计。
2．图像的预处理。在准确采集每一帧图像的基础上，通过分析比较，运用自适应的邻域平均法滤除了图像噪声；分析了矩量保持、最小误差、模糊集和最大熵四种自适应的图像阈值分割算法，选用最小误差法较好地实现了粮虫图像的自适应二值化分割；提出运用种子点火的连通区域标定法对多目标的粮虫二值化图像进行目标标定，解决了多目标害虫的识别问题。
3．特征提取。研究了主要储粮害虫的形态特征，提取了粮仓中长角扁谷盗、锯谷盗、玉米象、杂拟谷盗、赤拟谷盗、大谷盗等12种9类常见储粮害虫的面积、周长、复杂度、不变矩等14个形态特征；运用K-L变换将原始特征向量有效地压缩为6维。
4．BP神经网络分类器的设计。针对传统BP算法易陷入局部极小和收敛速度慢的问题，分析了各种改进方法，提出了新的联合优化算法，使网络的收敛速度大大加快；研究了奇异值分析、线性相关性分析、输
郑州大学硕士学位论文 摘要
出影响因素分析和自适应线性单元学习四种网络结构优化方法，实现了
BP神经网络隐层节点的优化；所设计的BP网络分类器对9类害虫的识
别率达到了95石％。
5．RBF神经网络分类器的设计。分析了现有的RBF神经网络训练
方法，选用模糊C均值聚类法进行网络隐层参数的学习，用递推最小二
廖，乘算法进行输出层权值的学习，所训练的网络对9类害虫的识别率达到
瞪 了 96．7％。
6．模糊和模糊BP神经网络分类器的设计。结合本对象，研究分析
了＊。uss模糊分类器的不足，提出将模糊技术和神经网络技术相结合来
实现储粮害虫分类的思想，设计了模糊BP神经网络分类器，其识别率
比hu％模糊分类器提高了15．6％，达到95石％：分析比较了所设计的
四种分类器，选用＊＊F网络分类器进行了现场模拟实验，对9类害虫的
在线识别率达到86．5％。
7．用VC＋＋6刀开发了基于图像识别的储粮害虫检测系统软件，并
建立了图像处理和分类器设计算法的动态连接库。
8．实现了储粮害虫的在线检测，为进一步实现整个系统的产品化奠
定了良好的基础。
本文用图像处理、模式识别、神经网络等技术来实现储粮害虫的在
线检测，设计了图像识别储粮害虫检测系统，重点研究了储粮害虫分类
器的设计。本系统的实现，可有助于粮库管理人员科学地选择决策依据，
并及时采取合理的措施，达到粮食保质、保鲜、保量的目的；并且基于
图像识别的害虫检测方法具有无污染、效率高、成本低、可以利用并完
善粮库己有的计算机粮情测控系统等优点，因此，本课题的研究具有广
。阔的应用前景和实用价值。
目前，本系统在河南郑州。民权等国家粮食储备库的现场实验，己
能以86．5％的识别率在线检测出粮仓中常见的12种害虫，得到了粮食储
藏方面专家的一致好评。不过，系统目前还不能解诀幼虫的识别问题；
我国常见的储粮害虫有 60多种，而系统仅能识别出 12种，为了更好地
满足储粮害虫检测的需要，实现系统的产品化，仍然需要进一步的研究
与探讨。
This thesis is sponsored by the Natural Science Foundation of Henan Province and the Opening Foundation of National Laboratory of Pattern Recognition, CAS.
How to control the stored-grain insects is the research hot of foodstuff storage trade in the world because the losses caused by stored-grain insects are very serious during the course of foodstuff storage. In 1972, the Organization of International Foodstuff brought forward Integrated Pest Management based on ecology. And the insect detection is the important component of the IPM. Relying on accurately detecting stored-grain insects can objectively manage insects, decrease losses, lower waste caused by blinded management and relieve the pollution to grain and environment.
Now, our country mainly uses sampling method to detect stored-grain insects, which has several drawbacks: the large working capacity, the low efficiency and the easy influence by the detecting workers. So it is difficult to meet the work needs of foodstuff storage. To follow the modern development of foodstuff storage technology, the paper gives a new method to detect stored-grain insects on-line based on image recognition. Center around this subject, the main work has been done as follows:
1. The software and hardware design of the stored-grain insects detection system based on image recognition. Has analyzed and compared several important methods of stored-grain insects detection, and describes the significance of studying the on-line stored-grain insects detection based on image recognition; and also gives the detailed detecting scheme and the software and hardware realization.
2. Image acquisition and pre-processing. Based on the exactly gathering a frame of image, after analysis and comparison, using adaptive neighborhood mean method smoothes the image; then makes a study of minimum-error, moments-preserving, fussy sets and maximum entropy these four adaptive threshold methods, accurately segments the image by the method of minimum-error threshold; By use of grass-label method labels the different multi- insect objects and solves the problem of the classification of
multi-insect objects.
3. Features extraction. Studies shape features of most important insects, extracts fourteen features (area, perimeter, complexity, invariant moments et. al) of twelve kinds, nine categories of insects: Cryptolestes minutus (Oliver), Oryzaephilus surinamensis, Sitophilus zeamais, T.confusum Jacguelin, Tribolium castaneum(Herbst), Tenebroides mauritanicus (Linnaeus), et al. And then using K-L transform reduces the dimension of original feature vector to six.
4. BP neural networks classifier design. In order to solve the conventional BP algorithm shortcomings of the easy falling into local-minimization and the slow convergence, has studied several improving methods, and comes up with a new combined improving method, which makes the learning speed much faster; makes a study of four optimizing networks' structure methods of SVD analyzing, linear relativity analyzing, output-influencing factor analyzing and adaptive linear unit learning, and optimizes the hidden nets of BP neural networks; the accurate recognition ratio of the designed BP neural networks classifier reaches 95.6 percent;
5. RBF neural networks classifier design. Analyses the existing methods of training RBF neural networks, and chooses fussy c-mean clustering method to determine the hidden nets parameters and RLS method to train the weights of output net-units; and so-trained RBF networks is able to accurately recognition nine categories of insects by the ratio of 96.7 percent;
6. Fussy and fussy BP neural networks classifier design. Based on our research subject-stored-grain insects, has analyzed the shortcomings of Gauss fussy classifier, and puts forward the thought of combining the technology of neural networks and fussy sets to recognize stored-grain insects; So designs a fussy BP neural networks classifier, whose accurate re
在粮食储藏期间，因害虫危害所造成的损失十分严重，怎样有效防治储粮害虫，一直都是世界各国粮食储藏行业的研究热点。早在1972年，联合国粮农组织就提出了建立在生态学基础上的害虫综合防治策略(Integrated Pests Management)，而害虫的检测是IPM体系的重要组成部分。只有准确的检测，才能做到有目的的防治，既不会因虫害造成损失，也不会因盲目防治造成浪费，加重对粮食、环境的污染。
目前，我国主要采用人工扦样的方法来检测储粮害虫，由于扦样法技术落后、工作量大、效率低、还受工作人员主观因素的影响，因此难以适应我国粮食储藏工作的需要。为适应我国储粮现代化建设的需要，我们提出了基于图像识别的在线检测储粮害虫的新思路，围绕该课题本文主要进行了以下几个方面的研究工作：
1．图像识别储粮害虫检测系统的软硬件设计。分析比较了现有几种主要的储粮害虫检测方法，指出进行储粮害虫图像识别在线检测研究的意义；给出了检测系统的具体方案，软件和硬件的设计。
2．图像的预处理。在准确采集每一帧图像的基础上，通过分析比较，运用自适应的邻域平均法滤除了图像噪声；分析了矩量保持、最小误差、模糊集和最大熵四种自适应的图像阈值分割算法，选用最小误差法较好地实现了粮虫图像的自适应二值化分割；提出运用种子点火的连通区域标定法对多目标的粮虫二值化图像进行目标标定，解决了多目标害虫的识别问题。
3．特征提取。研究了主要储粮害虫的形态特征，提取了粮仓中长角扁谷盗、锯谷盗、玉米象、杂拟谷盗、赤拟谷盗、大谷盗等12种9类常见储粮害虫的面积、周长、复杂度、不变矩等14个形态特征；运用K-L变换将原始特征向量有效地压缩为6维。
4．BP神经网络分类器的设计。针对传统BP算法易陷入局部极小和收敛速度慢的问题，分析了各种改进方法，提出了新的联合优化算法，使网络的收敛速度大大加快；研究了奇异值分析、线性相关性分析、输
郑州大学硕士学位论文 摘要
出影响因素分析和自适应线性单元学习四种网络结构优化方法，实现了
BP神经网络隐层节点的优化；所设计的BP网络分类器对9类害虫的识
别率达到了95石％。
5．RBF神经网络分类器的设计。分析了现有的RBF神经网络训练
方法，选用模糊C均值聚类法进行网络隐层参数的学习，用递推最小二
廖，乘算法进行输出层权值的学习，所训练的网络对9类害虫的识别率达到
瞪 了 96．7％。
6．模糊和模糊BP神经网络分类器的设计。结合本对象，研究分析
了＊。uss模糊分类器的不足，提出将模糊技术和神经网络技术相结合来
实现储粮害虫分类的思想，设计了模糊BP神经网络分类器，其识别率
比hu％模糊分类器提高了15．6％，达到95石％：分析比较了所设计的
四种分类器，选用＊＊F网络分类器进行了现场模拟实验，对9类害虫的
在线识别率达到86．5％。
7．用VC＋＋6刀开发了基于图像识别的储粮害虫检测系统软件，并
建立了图像处理和分类器设计算法的动态连接库。
8．实现了储粮害虫的在线检测，为进一步实现整个系统的产品化奠
定了良好的基础。
本文用图像处理、模式识别、神经网络等技术来实现储粮害虫的在
线检测，设计了图像识别储粮害虫检测系统，重点研究了储粮害虫分类
器的设计。本系统的实现，可有助于粮库管理人员科学地选择决策依据，
并及时采取合理的措施，达到粮食保质、保鲜、保量的目的；并且基于
图像识别的害虫检测方法具有无污染、效率高、成本低、可以利用并完
善粮库己有的计算机粮情测控系统等优点，因此，本课题的研究具有广
。阔的应用前景和实用价值。
目前，本系统在河南郑州。民权等国家粮食储备库的现场实验，己
能以86．5％的识别率在线检测出粮仓中常见的12种害虫，得到了粮食储
藏方面专家的一致好评。不过，系统目前还不能解诀幼虫的识别问题；
我国常见的储粮害虫有 60多种，而系统仅能识别出 12种，为了更好地
满足储粮害虫检测的需要，实现系统的产品化，仍然需要进一步的研究
与探讨。
This thesis is sponsored by the Natural Science Foundation of Henan Province and the Opening Foundation of National Laboratory of Pattern Recognition, CAS.
How to control the stored-grain insects is the research hot of foodstuff storage trade in the world because the losses caused by stored-grain insects are very serious during the course of foodstuff storage. In 1972, the Organization of International Foodstuff brought forward Integrated Pest Management based on ecology. And the insect detection is the important component of the IPM. Relying on accurately detecting stored-grain insects can objectively manage insects, decrease losses, lower waste caused by blinded management and relieve the pollution to grain and environment.
Now, our country mainly uses sampling method to detect stored-grain insects, which has several drawbacks: the large working capacity, the low efficiency and the easy influence by the detecting workers. So it is difficult to meet the work needs of foodstuff storage. To follow the modern development of foodstuff storage technology, the paper gives a new method to detect stored-grain insects on-line based on image recognition. Center around this subject, the main work has been done as follows:
1. The software and hardware design of the stored-grain insects detection system based on image recognition. Has analyzed and compared several important methods of stored-grain insects detection, and describes the significance of studying the on-line stored-grain insects detection based on image recognition; and also gives the detailed detecting scheme and the software and hardware realization.
2. Image acquisition and pre-processing. Based on the exactly gathering a frame of image, after analysis and comparison, using adaptive neighborhood mean method smoothes the image; then makes a study of minimum-error, moments-preserving, fussy sets and maximum entropy these four adaptive threshold methods, accurately segments the image by the method of minimum-error threshold; By use of grass-label method labels the different multi- insect objects and solves the problem of the classification of
multi-insect objects.
3. Features extraction. Studies shape features of most important insects, extracts fourteen features (area, perimeter, complexity, invariant moments et. al) of twelve kinds, nine categories of insects: Cryptolestes minutus (Oliver), Oryzaephilus surinamensis, Sitophilus zeamais, T.confusum Jacguelin, Tribolium castaneum(Herbst), Tenebroides mauritanicus (Linnaeus), et al. And then using K-L transform reduces the dimension of original feature vector to six.
4. BP neural networks classifier design. In order to solve the conventional BP algorithm shortcomings of the easy falling into local-minimization and the slow convergence, has studied several improving methods, and comes up with a new combined improving method, which makes the learning speed much faster; makes a study of four optimizing networks' structure methods of SVD analyzing, linear relativity analyzing, output-influencing factor analyzing and adaptive linear unit learning, and optimizes the hidden nets of BP neural networks; the accurate recognition ratio of the designed BP neural networks classifier reaches 95.6 percent;
5. RBF neural networks classifier design. Analyses the existing methods of training RBF neural networks, and chooses fussy c-mean clustering method to determine the hidden nets parameters and RLS method to train the weights of output net-units; and so-trained RBF networks is able to accurately recognition nine categories of insects by the ratio of 96.7 percent;
6. Fussy and fussy BP neural networks classifier design. Based on our research subject-stored-grain insects, has analyzed the shortcomings of Gauss fussy classifier, and puts forward the thought of combining the technology of neural networks and fussy sets to recognize stored-grain insects; So designs a fussy BP neural networks classifier, whose accurate re
引文
[1] 付昌斌，张兴．利用植物性物质防治储粮害虫研究进展[J]．粮食储藏，1996，25(5)：6-10．
[2] 姚渭，孟书伟．新仓型储粮害虫陷阱检测技术研究及应用[J]．粮食储藏，2000，30(5)：3-8．
[3] 陈启宗，黄建国．仓库昆虫图册[M]．北京：科学出版社，1985．
[4] 国家粮食局，“十五”粮食行业科技发展规划，2001．
[5] 张国梁．对储粮害虫化学防治现状的探讨[J]．粮食储藏，1996，25(2)：14-15．
[6] 徐永安，王建业等．磷化氢仓外投药环流熏蒸新技术成套装备产业化开发研究[J]．粮食储藏，1999，28(2)：3-9．
[7] 张国梁．储粮害虫防护剂使用现状与展望[J]．粮食储藏，1996，4(25)：27-31．
[8] 沈兆鹏．储粮害虫防治新方法——生物防治法[J]．粮食储藏，1997，26(2)：26-30．
[9] 梁永生，储粮害虫防治策略——害虫综合治理[J]．粮油仓储科技通讯，1994(4)：7-11．
[10] 姚渭等．粮虫陷阱检测技术研究及应用[A]．中国粮油学会首届学术年会论文选集(储藏专业卷)[C]．石家庄，2000：7-16．
[11] Rees D. P.. Review of the Response of Stored Product Insects to Light of Various Wavelengths, with Particular Reference to the Design and Use of Light Traps for Population Monitoring [J]. Trop. Sci., 1985, 25: 197-213.
[12] Hagstrum D. W., et al. Predicting Insect Density from Probe Trap Catch in Farm-stored Wheat[J]. Stored Prod. Res. 1998, 34: 251-262.
[13] 赵奇，田本志．国外用人工合成信息素防治仓储害虫的研究[J]．中国粮油学报，1999,14(2)：52-55．
[14] 蒋小龙．仓储害虫诱捕器种类综述[J]．粮食储藏，1992，21(2)：10-15．
[15] Javer A., Borden J. H.,et al. Evaluation of Pheromone - baited Traps for Monitoring of Cucujid and Tenebrionid Beetles in Stored Grain[J].Econ. Entomol, 1990, 83(1): 268-272.
[16] Michael. A. M. Development of a Pheromone Trap for Monitoring Tribolium Castaneum[J]. Stored Prod. Res, 1992,28(4):245-249.
[17] R. W., Shuman E., et al. Acoustic counting of adult insects with different rates and intensities of sound production in stored wheat[J]. Econ. Entomol. 1997, 90:1032-1038.
[18] Chambers J., et al. NIR Analysis for the Detection of Insect Pests in Cereal Grains, Proc[C]. Int. Conf. on Diffuse Spectroscopy, MD USA, 1992: 96-100.
[19] Flinn, P. W., Hagstrum W. E., et al. Effects of time of aeration, bin size, and latitude on insect populations in strored wheat: a simulation study[J]. Econ. Entomol. 1997, 90: 646-651.
[20]王进军，赵志模等．嗜卷书虱的实验生态研究[J]．昆虫学报．1999，42(3)：277-282．
[21]李庆中，汪懋华．农业生物模式识别中的计算机视觉技术[J]．中国图象图形学学报，1999，4(7)：610-616．
[22]宁纪锋，何东等．基于图像形态学的球形果实品质检测研究[J]．农机化研究，2001，(3)：28-47．
[23] Slaughter D. C., Harrell R. C.. Discriminating Fruit for Robotic Harvest using Color in Natural Outdoor Scenes[J]. Trans. Of the ASAE, 1989, 32(2): 757-763.
[24] Shimizu H., Heins R. D.. Computer; vision- based System for Plant Growth Analysis[J]. Trans. of the ASAE, 1995, 38(3): 959-964.
[25]梁永生．美国粮食储藏技术研究近况[J]．粮食储藏，1994，23(1)：31-40．
[26] Majumdar S., Jayas D. S.. Classification of Bulk Samples of Cereal Grains using Machine Vision[J]. Journal of Agriculture Engineering Research, 1999, 73:35-47.
[27] Liu J., Paulsen R. M.. Corn Whiteness Measurement and Classification using Machine Vision[J]. Trans. of the ASAE, 2000, 43(3):757-763.
[28] Zayas I. Y., Inna Y., et al., Detection of Insects in Bulk Wheat Samples with Machine Vision[J]. Trans. Of the ASAE., 1998 (3): 883-888.
[29]唐为民，呼玉山．近十年我国粮食仓储技术研究进展及推广应用状况[J]．粮食储藏，1997，26(4)：15-21．
[30]徐昉，邱道尹等．粮仓害虫的特征提取与分类研究[J]．郑州工业大学学报，2000，21(4)：62-65．
[31]徐昉，邱道尹等．图像识别在粮仓害虫检测方面的应用研究[J]．郑州工程学院学报，2001，22(1)：78-81．
[32]王明洁．对中国储粮区域的研究[J]．郑州粮食学院学报，2000，21(1)：62-66．
[33]周长发．精通Visual C++图像编程[M]．北京：电子工业出版社，2000．
[34]王德民，王庆麟．一种新的图像平滑方法及其性能评价[J]．自动化学报，1986，12(4)：354-359．
[35]吴一全，朱兆达．图像处理中阈值选取方法30年(1962-1992)的进展(一)[J]．数据采集与处理，1993，8(3)：193-201．
[36]吴一全，朱兆达．图像处理中阈值选取方法30年(1962-1992)的进展(二)[J]．数据采集与处理，1993，8(4)：268-278．
[37] Huang L K, Wang M J. Image thresholding by minimizing the measures of fuzziness[J].Pattern Recognition, 1995,28(1): 41-51.
[38]徐昉．基于图像识别的储粮害虫检测系统的研究[D]．郑州大学硕士学位论文，2001．5．
[39]崔屹．图像处理技术与应用[M]．北京：电子工业出版社，1997．
[40]Kenneth R Castleman．数字图像处理[M]．北京：电子工业出版社，1998．
[41]边肇祺．模式识别[M]．北京：清华大学出版社，2000．
[42] Minichi, Kudo, Jack Sklansky. Comparison of algorithms that select features for pattern classification[J]. Pattern Recognition, 2000, 33(1 ):25-41.
[43]黄德双．神经网络模式识别系统理论[M]．北京：电子工业出版社，1996．
[44] Widrow B, et al. Neural networks application in industry, business and science[J].Communication of the ACM, 1994,37:93-105.
[45]阎平凡，张长水等．人工神经网络与模拟进化计算[M]．北京：清华大学出版社，2000．
[46] Funanashi Kolmogorov. On the approximate realization of continuous mappings by neural networks[J]. Neural Networks, 1989,2:359-366.
[47]张铃，张钹．神经网络中BP算法的分析[J]．模式识别与人工智能，1994，7(3)：191-195．
[48]王科俊，李国斌．几种变学习率的快速BP算法比较研究[J]．哈尔滨工程大学学报，1997，18(3)：31，35，
[49]邓志东，孙增圻．利用线性再励的自适应变步长快速BP算法[J]．模式识别与人工智能，1993,6(4)：319-323．
[50] Jacobs R A. Increased rates of convergence through learning rate adaptation, Neural Networks, 1988,1:295-307.
[51]李平，刘炯等．前馈神经网络的结构与范化[J]．长春光学机械学院学报，1997,20(2)：63-67．
[52]高大启．有教师地线性基本函数前向三层神经网络结构研究[J]．计算机学报，1998,21(1)：80-86．
[53]周东华，叶银忠著．现代故障诊断与容错控制[M]．北京：清华大学出版社，2000．
[54]滕召胜，罗隆福等．智能检测系统与数据融合[M]．北京：机械工业出版社，2000．
[55]金丕彦，芮勇．BP各种改进算法的研究及应用[J]．南京航空航天大学学报，1994,26(增)：201-203．
[56]方剑，席裕庚。神经网络结构设计的准则和方法[J] 信息与控制，1996,25(3)：156-164
[57]何述东，瞿坦等．多层前向神经网络结构的研究进展[J]．控制理论与应用，1998,15(3)：313-319．
[58]Bartle EricB. Dynamic node architecture learning: an information theoretic approach[J]. Neural Network, 1994,7:29-140.
[59]Murata J. A structure designing method for feedforward and recurrent neural networks based on a net significance measure[C]. Proc. Of 4th IEE Int. Conf. on ANN, 1995,358-363.
[60]Angeline Peter J, et al. An Evolutionary Algorithm that constructs recurrent neural networks[J].IEEE Trans on Neural Networks, 1994,5(1):54-65
[61]刘光中，李晓峰．人工神经网络BP算法的改进和结构的自调整[J]．运筹学学报，2001，5(1)：82-88．
[62]Psichogios D. C., et al. SVD-NET: An algorithm that automatically selects network structure[J]. IEEE Trans. Neural Networks, 1994,5(3):513-515.
[63]吕伯权，李天怿．一种优化神经网络结构方法[J]．华北电力大学学报，1996,23(3)：36-40．
[64]Gao Da-qi,Wu Shou-yi. An optimization method for the topological structure of feed-forward multiplayer neural networks[J]. Pattern Recognition, 1998,31(9):1337-1342.
[65]陈涛，蒋林等．基于正交最小二乘算法的径向基函数网络设计[J]．中国机械工程，1997，8(6)：95-97．
[66]Chen S, et al. Orthogonal least square learning algorithm for RBF networks[J]. IEEE Trans NN, 1991,2:302-309.
[67]Guosui Liu, et al. A novel neural network[J]. IJCNNSP, 1995,1:13-16.
[68]Zadeh L.A.. Fussy sets[J]. Information and Control, 1965,8:338-335.
[69]文贡坚，王润生．基于模糊决策的快速识别多类目标的方法[J]．模式识别与人工智能，1997,10(2)：106-111．
[2] 姚渭，孟书伟．新仓型储粮害虫陷阱检测技术研究及应用[J]．粮食储藏，2000，30(5)：3-8．
[3] 陈启宗，黄建国．仓库昆虫图册[M]．北京：科学出版社，1985．
[4] 国家粮食局，“十五”粮食行业科技发展规划，2001．
[5] 张国梁．对储粮害虫化学防治现状的探讨[J]．粮食储藏，1996，25(2)：14-15．
[6] 徐永安，王建业等．磷化氢仓外投药环流熏蒸新技术成套装备产业化开发研究[J]．粮食储藏，1999，28(2)：3-9．
[7] 张国梁．储粮害虫防护剂使用现状与展望[J]．粮食储藏，1996，4(25)：27-31．
[8] 沈兆鹏．储粮害虫防治新方法——生物防治法[J]．粮食储藏，1997，26(2)：26-30．
[9] 梁永生，储粮害虫防治策略——害虫综合治理[J]．粮油仓储科技通讯，1994(4)：7-11．
[10] 姚渭等．粮虫陷阱检测技术研究及应用[A]．中国粮油学会首届学术年会论文选集(储藏专业卷)[C]．石家庄，2000：7-16．
[11] Rees D. P.. Review of the Response of Stored Product Insects to Light of Various Wavelengths, with Particular Reference to the Design and Use of Light Traps for Population Monitoring [J]. Trop. Sci., 1985, 25: 197-213.
[12] Hagstrum D. W., et al. Predicting Insect Density from Probe Trap Catch in Farm-stored Wheat[J]. Stored Prod. Res. 1998, 34: 251-262.
[13] 赵奇，田本志．国外用人工合成信息素防治仓储害虫的研究[J]．中国粮油学报，1999,14(2)：52-55．
[14] 蒋小龙．仓储害虫诱捕器种类综述[J]．粮食储藏，1992，21(2)：10-15．
[15] Javer A., Borden J. H.,et al. Evaluation of Pheromone - baited Traps for Monitoring of Cucujid and Tenebrionid Beetles in Stored Grain[J].Econ. Entomol, 1990, 83(1): 268-272.
[16] Michael. A. M. Development of a Pheromone Trap for Monitoring Tribolium Castaneum[J]. Stored Prod. Res, 1992,28(4):245-249.
[17] R. W., Shuman E., et al. Acoustic counting of adult insects with different rates and intensities of sound production in stored wheat[J]. Econ. Entomol. 1997, 90:1032-1038.
[18] Chambers J., et al. NIR Analysis for the Detection of Insect Pests in Cereal Grains, Proc[C]. Int. Conf. on Diffuse Spectroscopy, MD USA, 1992: 96-100.
[19] Flinn, P. W., Hagstrum W. E., et al. Effects of time of aeration, bin size, and latitude on insect populations in strored wheat: a simulation study[J]. Econ. Entomol. 1997, 90: 646-651.
[20]王进军，赵志模等．嗜卷书虱的实验生态研究[J]．昆虫学报．1999，42(3)：277-282．
[21]李庆中，汪懋华．农业生物模式识别中的计算机视觉技术[J]．中国图象图形学学报，1999，4(7)：610-616．
[22]宁纪锋，何东等．基于图像形态学的球形果实品质检测研究[J]．农机化研究，2001，(3)：28-47．
[23] Slaughter D. C., Harrell R. C.. Discriminating Fruit for Robotic Harvest using Color in Natural Outdoor Scenes[J]. Trans. Of the ASAE, 1989, 32(2): 757-763.
[24] Shimizu H., Heins R. D.. Computer; vision- based System for Plant Growth Analysis[J]. Trans. of the ASAE, 1995, 38(3): 959-964.
[25]梁永生．美国粮食储藏技术研究近况[J]．粮食储藏，1994，23(1)：31-40．
[26] Majumdar S., Jayas D. S.. Classification of Bulk Samples of Cereal Grains using Machine Vision[J]. Journal of Agriculture Engineering Research, 1999, 73:35-47.
[27] Liu J., Paulsen R. M.. Corn Whiteness Measurement and Classification using Machine Vision[J]. Trans. of the ASAE, 2000, 43(3):757-763.
[28] Zayas I. Y., Inna Y., et al., Detection of Insects in Bulk Wheat Samples with Machine Vision[J]. Trans. Of the ASAE., 1998 (3): 883-888.
[29]唐为民，呼玉山．近十年我国粮食仓储技术研究进展及推广应用状况[J]．粮食储藏，1997，26(4)：15-21．
[30]徐昉，邱道尹等．粮仓害虫的特征提取与分类研究[J]．郑州工业大学学报，2000，21(4)：62-65．
[31]徐昉，邱道尹等．图像识别在粮仓害虫检测方面的应用研究[J]．郑州工程学院学报，2001，22(1)：78-81．
[32]王明洁．对中国储粮区域的研究[J]．郑州粮食学院学报，2000，21(1)：62-66．
[33]周长发．精通Visual C++图像编程[M]．北京：电子工业出版社，2000．
[34]王德民，王庆麟．一种新的图像平滑方法及其性能评价[J]．自动化学报，1986，12(4)：354-359．
[35]吴一全，朱兆达．图像处理中阈值选取方法30年(1962-1992)的进展(一)[J]．数据采集与处理，1993，8(3)：193-201．
[36]吴一全，朱兆达．图像处理中阈值选取方法30年(1962-1992)的进展(二)[J]．数据采集与处理，1993，8(4)：268-278．
[37] Huang L K, Wang M J. Image thresholding by minimizing the measures of fuzziness[J].Pattern Recognition, 1995,28(1): 41-51.
[38]徐昉．基于图像识别的储粮害虫检测系统的研究[D]．郑州大学硕士学位论文，2001．5．
[39]崔屹．图像处理技术与应用[M]．北京：电子工业出版社，1997．
[40]Kenneth R Castleman．数字图像处理[M]．北京：电子工业出版社，1998．
[41]边肇祺．模式识别[M]．北京：清华大学出版社，2000．
[42] Minichi, Kudo, Jack Sklansky. Comparison of algorithms that select features for pattern classification[J]. Pattern Recognition, 2000, 33(1 ):25-41.
[43]黄德双．神经网络模式识别系统理论[M]．北京：电子工业出版社，1996．
[44] Widrow B, et al. Neural networks application in industry, business and science[J].Communication of the ACM, 1994,37:93-105.
[45]阎平凡，张长水等．人工神经网络与模拟进化计算[M]．北京：清华大学出版社，2000．
[46] Funanashi Kolmogorov. On the approximate realization of continuous mappings by neural networks[J]. Neural Networks, 1989,2:359-366.
[47]张铃，张钹．神经网络中BP算法的分析[J]．模式识别与人工智能，1994，7(3)：191-195．
[48]王科俊，李国斌．几种变学习率的快速BP算法比较研究[J]．哈尔滨工程大学学报，1997，18(3)：31，35，
[49]邓志东，孙增圻．利用线性再励的自适应变步长快速BP算法[J]．模式识别与人工智能，1993,6(4)：319-323．
[50] Jacobs R A. Increased rates of convergence through learning rate adaptation, Neural Networks, 1988,1:295-307.
[51]李平，刘炯等．前馈神经网络的结构与范化[J]．长春光学机械学院学报，1997,20(2)：63-67．
[52]高大启．有教师地线性基本函数前向三层神经网络结构研究[J]．计算机学报，1998,21(1)：80-86．
[53]周东华，叶银忠著．现代故障诊断与容错控制[M]．北京：清华大学出版社，2000．
[54]滕召胜，罗隆福等．智能检测系统与数据融合[M]．北京：机械工业出版社，2000．
[55]金丕彦，芮勇．BP各种改进算法的研究及应用[J]．南京航空航天大学学报，1994,26(增)：201-203．
[56]方剑，席裕庚。神经网络结构设计的准则和方法[J] 信息与控制，1996,25(3)：156-164
[57]何述东，瞿坦等．多层前向神经网络结构的研究进展[J]．控制理论与应用，1998,15(3)：313-319．
[58]Bartle EricB. Dynamic node architecture learning: an information theoretic approach[J]. Neural Network, 1994,7:29-140.
[59]Murata J. A structure designing method for feedforward and recurrent neural networks based on a net significance measure[C]. Proc. Of 4th IEE Int. Conf. on ANN, 1995,358-363.
[60]Angeline Peter J, et al. An Evolutionary Algorithm that constructs recurrent neural networks[J].IEEE Trans on Neural Networks, 1994,5(1):54-65
[61]刘光中，李晓峰．人工神经网络BP算法的改进和结构的自调整[J]．运筹学学报，2001，5(1)：82-88．
[62]Psichogios D. C., et al. SVD-NET: An algorithm that automatically selects network structure[J]. IEEE Trans. Neural Networks, 1994,5(3):513-515.
[63]吕伯权，李天怿．一种优化神经网络结构方法[J]．华北电力大学学报，1996,23(3)：36-40．
[64]Gao Da-qi,Wu Shou-yi. An optimization method for the topological structure of feed-forward multiplayer neural networks[J]. Pattern Recognition, 1998,31(9):1337-1342.
[65]陈涛，蒋林等．基于正交最小二乘算法的径向基函数网络设计[J]．中国机械工程，1997，8(6)：95-97．
[66]Chen S, et al. Orthogonal least square learning algorithm for RBF networks[J]. IEEE Trans NN, 1991,2:302-309.
[67]Guosui Liu, et al. A novel neural network[J]. IJCNNSP, 1995,1:13-16.
[68]Zadeh L.A.. Fussy sets[J]. Information and Control, 1965,8:338-335.
[69]文贡坚，王润生．基于模糊决策的快速识别多类目标的方法[J]．模式识别与人工智能，1997,10(2)：106-111．