农业工程学报
農業工程學報
농업공정학보
2015年
4期
325-331
,共7页
田有文%程怡%王小奇%刘思伽
田有文%程怡%王小奇%劉思伽
전유문%정이%왕소기%류사가
无损检测%图像处理%主成分分析%苹果虫伤%果梗/花萼%高光谱成像%支持向量机
無損檢測%圖像處理%主成分分析%蘋果蟲傷%果梗/花萼%高光譜成像%支持嚮量機
무손검측%도상처리%주성분분석%평과충상%과경/화악%고광보성상%지지향량궤
nondestructive examination%image processing%principal component analysis%apple insect damage%stem/calyx%hyperspectral imaging%support vector machine
为了快速、准确、无损检测在果梗/花萼的干扰下苹果虫伤缺陷,该文利用高光谱成像技术,首先选取正常果和虫伤果各80个,提取并分析了苹果表面感兴趣区域(虫伤区域、果梗区域、花萼区域、正常区域)的光谱曲线,结合824 nm波长特征图像的阈值分割和主成分分析,对获得的第一主成分图像提取160×120像素大小的感兴趣区域。然后提取感兴趣区域的能量、熵、惯性矩和相关性4个纹理特征,融合646、824 nm波段的相对光谱反射率的光谱特征,采用支持向量机对苹果虫伤区域和正常区域、果梗/花萼区域进行识别。试验结果表明:选取160×120像素大小的感兴趣区域图像、采用径向基核函数对正常果、果梗/花萼果与虫伤果的识别效果最好,总体识别率为97.8%。该研究为苹果质量等级在线评判提供理论依据。
為瞭快速、準確、無損檢測在果梗/花萼的榦擾下蘋果蟲傷缺陷,該文利用高光譜成像技術,首先選取正常果和蟲傷果各80箇,提取併分析瞭蘋果錶麵感興趣區域(蟲傷區域、果梗區域、花萼區域、正常區域)的光譜麯線,結閤824 nm波長特徵圖像的閾值分割和主成分分析,對穫得的第一主成分圖像提取160×120像素大小的感興趣區域。然後提取感興趣區域的能量、熵、慣性矩和相關性4箇紋理特徵,融閤646、824 nm波段的相對光譜反射率的光譜特徵,採用支持嚮量機對蘋果蟲傷區域和正常區域、果梗/花萼區域進行識彆。試驗結果錶明:選取160×120像素大小的感興趣區域圖像、採用徑嚮基覈函數對正常果、果梗/花萼果與蟲傷果的識彆效果最好,總體識彆率為97.8%。該研究為蘋果質量等級在線評判提供理論依據。
위료쾌속、준학、무손검측재과경/화악적간우하평과충상결함,해문이용고광보성상기술,수선선취정상과화충상과각80개,제취병분석료평과표면감흥취구역(충상구역、과경구역、화악구역、정상구역)적광보곡선,결합824 nm파장특정도상적역치분할화주성분분석,대획득적제일주성분도상제취160×120상소대소적감흥취구역。연후제취감흥취구역적능량、적、관성구화상관성4개문리특정,융합646、824 nm파단적상대광보반사솔적광보특정,채용지지향량궤대평과충상구역화정상구역、과경/화악구역진행식별。시험결과표명:선취160×120상소대소적감흥취구역도상、채용경향기핵함수대정상과、과경/화악과여충상과적식별효과최호,총체식별솔위97.8%。해연구위평과질량등급재선평판제공이론의거。
Insect damage is one of the main defects of the apple, which could make it lose edibility and greatly reduce the quality of the apple and commercial value. So whether apples exist insect damage is one of the important indicators of the grade apple quality. In this study, we aim to nondestructively detect insect damage on apples under the interference of stem / calyx with hyperspectral imaging technique. 160 ‘Red Fuji’ apples,including 80 intact and 80 insect infected, picked from an apple planting demonstration garden in the Shenbei New Area in Shenyang city. A hyperspectral imaging collection system with the wavelength range of 400-1 000 nm and spectral resolution ratio of 2.8 mm was established in order to acquire the hyperspectral images of these apple samples. These acquisition apples hyperspectral images were carried out black and white plate correction in order to eliminate the noise generated hyperspectral imaging instrument in the acquisition process. The extraction and analysis of spectral reflectance of apple surface interested region, which were insect damage region, stem/calyx region and normal region, exhibted great differences in spectral reflectance at the 824 nm wavelength. So, the images of the 824 nm wavelength were named the feature images. Then, the feature images were processed by threshold segmentation, dilation, and erosion operation. A binarization image was obtained in the end. The binarization image was used to mask for apple hyperspectral image in order to remove noise of background on hyperspectral image. So a mask apple hyperspectral image was obtained. These processed hyperspectral images were carried by principal component analysis. The optimum PC1 image was chosen and processed by the maximum entropy threshold segmentation to get the insect damage region, stem/calyx region. According to interested region segmentation results, an interested region image was obtained from the PC1 image of each sample, with 80×60 pixels, 160×120 pix and 240×180 pix, respectively. Later, there were 4 the texture features of the gray level co-occurrence matrix (of energy, entropy, moment of inertia and correlation) of the insect damage region, stem/calyx region and the normal region on apples of the PC1 image. In addition, whether the spectral relative reflectance of the apple surface insect damage region, stem/calyx region and normal regions, was visible or near infrared region, showing some differences. So the two spectral features of the spectrum relative reflectivity at 646 nm and 824 nm wavelength were chosen, which had larger relative reflectance differences between the apple surface insect damage region, stem/calyx region and normal regions in the visible region and near infrared region. For faster and more accurate detection of the apple insect damage, the texture features and the spectral feature vectors were merged as input of Support Vector Machine (SVM), which is a recognition method of insect damage on apples. Finally, via a comparative analysis of recognition results with differences interested region size among 80×60 pix, 160×120 pix and 240×180 pix, we gained a result that the recognition result of apple insect of 160×120 pix interested region was the best. Through a comparative analysis of recognition with differences kernel function of SVM among linear kernel, polynomial kernel, rbf kernel and sigmoid kernel. The recognition effect of radial basis kernel function was the best, with the overall recognition rate of 97.8%. The testing results showed that hyperspectral imaging technology can be used for identification of insect damage and stem/calyx on apple fruit with quick, accurate and non-destructive detection and provided a theoretical basis for subsequent developing online apple quality detecting and grading system based on multispectral imaging technique.