农业工程学报
農業工程學報
농업공정학보
2013年
24期
155-162
,共8页
王传宇%郭新宇%吴升%肖伯祥%杜建军
王傳宇%郭新宇%吳升%肖伯祥%杜建軍
왕전우%곽신우%오승%초백상%두건군
机器视觉%图像配准%图像融合%全景图%玉米果穗
機器視覺%圖像配準%圖像融閤%全景圖%玉米果穗
궤기시각%도상배준%도상융합%전경도%옥미과수
computer visions%image registration%image fusion%panoramas%maize ear
为了在利用图像技术无损考察玉米果穗形态指标时,能够利用一幅图像显示整个玉米果穗的外形,从而减少多幅图像拼接产生的重叠和处理不便,该文提出一种新的基于机器视觉的玉米果穗考种方法与配套装置,首先拍摄旋转玉米果穗图像序列,应用SIFT(scale invariant feature transform)算法获取图像特征点,对特征点随机采样计算单应矩阵并进行一致性检测排除外点,将前后2帧图像注册到同一坐标系。然后采用动态规划法寻找前后2帧拼接图像的缝合线,按缝合线切割图像,以图像模板高斯滤波权值融合缝合线两侧图像消除曝光差异。依次拼接、融合图像序列生成果穗全景图。对果穗全景图进行考种指标检测,试验结果表明:基于机器视觉的测量值与人工测量方式不存在显著性差异(显著水平α=0.05),该文所述方法可满足自动化考种的需求。
為瞭在利用圖像技術無損攷察玉米果穗形態指標時,能夠利用一幅圖像顯示整箇玉米果穗的外形,從而減少多幅圖像拼接產生的重疊和處理不便,該文提齣一種新的基于機器視覺的玉米果穗攷種方法與配套裝置,首先拍攝鏇轉玉米果穗圖像序列,應用SIFT(scale invariant feature transform)算法穫取圖像特徵點,對特徵點隨機採樣計算單應矩陣併進行一緻性檢測排除外點,將前後2幀圖像註冊到同一坐標繫。然後採用動態規劃法尋找前後2幀拼接圖像的縫閤線,按縫閤線切割圖像,以圖像模闆高斯濾波權值融閤縫閤線兩側圖像消除曝光差異。依次拼接、融閤圖像序列生成果穗全景圖。對果穗全景圖進行攷種指標檢測,試驗結果錶明:基于機器視覺的測量值與人工測量方式不存在顯著性差異(顯著水平α=0.05),該文所述方法可滿足自動化攷種的需求。
위료재이용도상기술무손고찰옥미과수형태지표시,능구이용일폭도상현시정개옥미과수적외형,종이감소다폭도상병접산생적중첩화처리불편,해문제출일충신적기우궤기시각적옥미과수고충방법여배투장치,수선박섭선전옥미과수도상서렬,응용SIFT(scale invariant feature transform)산법획취도상특정점,대특정점수궤채양계산단응구진병진행일치성검측배제외점,장전후2정도상주책도동일좌표계。연후채용동태규화법심조전후2정병접도상적봉합선,안봉합선절할도상,이도상모판고사려파권치융합봉합선량측도상소제폭광차이。의차병접、융합도상서렬생성과수전경도。대과수전경도진행고충지표검측,시험결과표명:기우궤기시각적측량치여인공측량방식불존재현저성차이(현저수평α=0.05),해문소술방법가만족자동화고충적수구。
Maize ear morphological characteristics have important applications in breeding, germplasm, and cultivation areas, subject to the extent of technology development in relevant areas, but the approach of surveying morphological characteristics is not highly automatic. In this paper, we present a new machine vision based method and a supporting device for maize ear morphological characteristic surveying. First, the maize ear was placed on a rotating component, which rotates the maize ear in a fixed angle interval in order to capture 16 images more or less. A preprocess was carried out of maize ear image sequences to remove the image background, and the remaining part of the maize ear image was passed to the next process. The SIFT (Scale Invariant Feature Transform) was used to extract image feature points, and the feature points in the neighboring images could be matched up according to SIFT feature points. The relative motion between the two images could be described by a homography, and an overdetermined equations composed of matching points and homography make specific values of homography available. Mismatched feature points will reduce the accuracy of the homography equation solution dramatically. We adopted a RANSAC (random sample consensus) method to remove the outlier of the matching points during the homography solving process. Secondly, according to the motion described by homography, the first image and the next image are registered to the same coordinate system, using the dynamic programming method to find the seam-line in the two images, cutting the redundancy region in the two images along the seam-line. Since the exposure of the two images had certain differences which led to image brightness near seam-line being slightly different, a weighted Gaussian filter was imposed on both sides of the stitching image to eliminate exposure difference. Finally, the fusion image according to the order in sequence generated the ear panorama, row number, number in a row, kernel number, and other parameters were extracted by processing the maize ear panorama. The experimental results showed that: there is no significant difference between the method proposed by this paper and manual measurement, and the method proposed can greatly strengthen the automation of the maize ear traits investigation.