浙江大学学报(农业与生命科学版)
浙江大學學報(農業與生命科學版)
절강대학학보(농업여생명과학판)
JOURNAL OF ZHEJIANG UNIVERSITY(AGRICULTURE & LIFE SCIENCES)
2014年
5期
585-590
,共6页
宋革联%韩瑞珍%张永华%何勇
宋革聯%韓瑞珍%張永華%何勇
송혁련%한서진%장영화%하용
农田害虫%自动识别%远程识别%图像处理
農田害蟲%自動識彆%遠程識彆%圖像處理
농전해충%자동식별%원정식별%도상처리
agricultural pests%automatic identification%remote identification%image processing
为了解决农业生产过程中虫害难以及时发现和识别的问题,开发了基于图像和无线传输技术的农田害虫远程自动识别系统.该系统包括1个主控平台和多个远程平台,主控平台与多个远程平台之间采用3G/4G无线网络通信,远端平台实时采集害虫图片,并将图片发送到主控端.主控端和远程端均具有获取图片、读入图片、特征提取、特征选择、专家识别、发送图片等功能模块.主控端和远端平台采用最大类间方差法将害虫从背景中分割出来,系统提取害虫的面积、周长、复杂度、偏心率和不变矩等16个形态学特征值,以及9个颜色特征值和基于灰度共生矩阵的10个纹理构成特征向量,并利用支持向量机分类器对农田中常见的稻纵卷叶螟、斜纹夜蛾、稻螟蛉、二化螟、玉米螟、白背飞虱和小地老虎等7种害虫进行分类.试验验证表明,系统对7种典型害虫的平均正确识别率为88.5%,取得了较好的效果.
為瞭解決農業生產過程中蟲害難以及時髮現和識彆的問題,開髮瞭基于圖像和無線傳輸技術的農田害蟲遠程自動識彆繫統.該繫統包括1箇主控平檯和多箇遠程平檯,主控平檯與多箇遠程平檯之間採用3G/4G無線網絡通信,遠耑平檯實時採集害蟲圖片,併將圖片髮送到主控耑.主控耑和遠程耑均具有穫取圖片、讀入圖片、特徵提取、特徵選擇、專傢識彆、髮送圖片等功能模塊.主控耑和遠耑平檯採用最大類間方差法將害蟲從揹景中分割齣來,繫統提取害蟲的麵積、週長、複雜度、偏心率和不變矩等16箇形態學特徵值,以及9箇顏色特徵值和基于灰度共生矩陣的10箇紋理構成特徵嚮量,併利用支持嚮量機分類器對農田中常見的稻縱捲葉螟、斜紋夜蛾、稻螟蛉、二化螟、玉米螟、白揹飛虱和小地老虎等7種害蟲進行分類.試驗驗證錶明,繫統對7種典型害蟲的平均正確識彆率為88.5%,取得瞭較好的效果.
위료해결농업생산과정중충해난이급시발현화식별적문제,개발료기우도상화무선전수기술적농전해충원정자동식별계통.해계통포괄1개주공평태화다개원정평태,주공평태여다개원정평태지간채용3G/4G무선망락통신,원단평태실시채집해충도편,병장도편발송도주공단.주공단화원정단균구유획취도편、독입도편、특정제취、특정선택、전가식별、발송도편등공능모괴.주공단화원단평태채용최대류간방차법장해충종배경중분할출래,계통제취해충적면적、주장、복잡도、편심솔화불변구등16개형태학특정치,이급9개안색특정치화기우회도공생구진적10개문리구성특정향량,병이용지지향량궤분류기대농전중상견적도종권협명、사문야아、도명령、이화명、옥미명、백배비슬화소지로호등7충해충진행분류.시험험증표명,계통대7충전형해충적평균정학식별솔위88.5%,취득료교호적효과.
Summary The automatic remote pest-identification system is developed aiming at solving the difficulty of detecting pests in good timing in farmland,which applies modern wireless image and data transaction tools.The system consists of one main controller and several supplemental remote platforms.In structure,every single platform includes function modules like image acquisition,image processing,feature extraction and classification, GPS and transmission module.These platforms communicate with each other through 3G/4G wireless network. Host end and the remote end has access to photos capture,pictures reading,feature extraction,feature selection, expert identification,pictures sending and so on.
<br> The workflow is as follows:1) the remote platforms capture the images of stationary pests with image acquisition system;2) the system extracts certain pests from the farming background according to the constructed vector including 1 6 morphological characteristic values such as area, perimeter, complexity, eccentricity and invariant moment,9 colors eigenvalues and 10 texture constituted feature;3) finally,the system classifies pests into 7 major types which are Cnaphalocrocis medinalis Guenee,Prodenia litura,Chilo suppressalis,Ostrinia nubilalis,Naranga aenesc,Sogatellafurcifera and Agrotis ypsilon Rottemberg with Otsu threshold segmentation method based on HSV color mode.The accuracy of this technology was proved to be 88.5%.
<br> The identification process can be completed in both remote platform and in the host control platform automatically after the pest images were compressed and transmitted to the host control platform through 3G/4G wireless network.With more and more application and information collected,the system will expand the sample library dynamically by saving the image into the local disks.
<br> The feasibility of the system is discussed and statistically significantly tested in the context.The advent of the automatic remote pest-identification system could help provide pest information about the farmland timely and accurately thus improving the prevention effect.