Machine vision system for online inspection of freshly slaughtered chickens

A machine vision system was developed and evaluated for the automation of online inspection to differentiate freshly slaughtered wholesome chickens from systemically diseased chickens. The system consisted of an electron-multiplying charge-coupled-device (EMCCD) camera used with an imaging spectrograph and controlled by a computer to obtain line-scan images quickly on a chicken processing line of a commercial poultry plant. The system scanned chicken carcasses on an eviscerating line operating at a speed of 140 chickens per minute. An algorithm was implemented in the system to automatically recognize individual carcasses entering and exiting the field of view, to locate the region of interest (ROI) of each chicken, to extract useful spectra from the ROI as inputs to the differentiation method, and to determine the condition for each carcass as being wholesome or systemically diseased. The system can acquire either hyperspectral or multispectral images without any cross-system calibration. The essential spectral features were selected from hyperspectral images of chicken samples. The differentiation of chickens on the processing line was then carried out using multispectral imaging. The high accuracy obtained from the evaluation results showed that the machine vision system can be applied successfully to automatic online inspection for chicken processing. Mention of trade names or commercial products is solely for the purpose of providing specific information and does not imply endorsement or recommendation by the USDA.     

基于自动子空间划分的高光谱数据特征提取

针对遥感高光谱图像数据量大、维数高的特点,提出了一种自动子空间划分方法用于高光谱图像数据量减小处理。该方法主要包括3个处理步骤:数据空间划分,子空间主成分分析和基于类别可分性准则的特征选择。该方法充分利用了高光谱图像各波段数据之间的局部相关性,将整个数据划分为若干个具有较强相关性的独立子空间,然后在子空间内利用主成分分析进行特征提取,根据各类地物间的类别可分性选择有效特征,最后利用地物分类来验证该方法的有效性。实验结果表明,该方法能够有效地实现高光谱图像数据维数减小和特征提取,同现有的自适应子空间分解方法和分段主成分变换方法相比,该方法所提取的特征用于分类时能获得较好的分类精度。利用该方法进行处理,当高光谱数据维数降低了90%时,9类地物分类实验的总体分类精度可以达到80.2%。     

Quality evaluation of pickling cucumbers using hyperspectral reflectance and transmittance imaging—Part II. Performance of a prototype

This paper reports on the development and evaluation of methods and algorithms for detecting both external and internal quality of pickling cucumbers using the hyperspectral reflectance and transmittance images acquired by an online prototype described in a previous paper [1]. Experiments were performed in 2 years on ‘Journey’ pickling cucumbers, some of which were subjected to mechanical stress to induce internal defect in seed cavity. Hyperspectral images of the ‘Journey’ pickling cucumbers were collected under reflectance, transmittance, and their combination modes. Partial least squares analysis was performed on spectra extracted from the hyperspectral images to predict firmness, color, and the presence of internal defect. The system performed well on predicting skin color (chroma and hue) with the coefficient of determination (R ²) ranging between 0.75 and 0.77; however, it had poor prediction of fruit firmness. Transmittance data in the spectral region of 675–1,000 nm provided the best detection of internal defect for the test pickling cucumbers, with the detection accuracy up to 99%. Up to the best four wavelength combinations were identified using linear discriminant analysis for internal defect detection. The hyperspectral imaging technique can be used for simultaneous detection of color, size, and internal defect on pickling cucumbers.     

高光谱遥感影像端元提取算法研究进展及分类

在给出端元的物理、代数和几何学解释基础上,对现有端元提取算法从算法设计机理出发,分为基于几何学、基于统计学和信号检测理论以及空间和光谱相结合三大类,并进一步对基于几何学的端元提取算法从技术处理手段差异细分为基于距离、体积、投影变换和最优化4种情况。结合端元提取算法分类,针对算法缺陷及改进思路,介绍了常见端元提取算法PPI、N-FINDR、UOSP、VCA、ICA、NMF和AMEE研究进展。最后,结合解混理论进展和工程应用实际,从技术综合和性能优化的角度指出了端元提取算法的研究展望。     

Towards universal broad leaf chlorophyll indices using PROSPECT simulated database and hyperspectral reflectance measurements

Fifty-three leaves were randomly sampled on different deciduous tree species, representing a wide range of chlorophyll contents, tree ages, and leaf structural features. Their reflectance was measured between 400 and 800 nm with a 1-nm step, and their chlorophyll content determined by extraction. A larger simulated database (11,583 spectra) was built using the PROSPECT model, in order to test, calibrate, and obtain universal indices, i.e., indices applicable to a wide range of species and leaf structure. To our knowledge, almost all leaf chlorophyll indices published in the literature since 1973 have been tested on both databases. Fourteen canonical types of indices (published ones and new ones) were identified, and their wavelengths calibrated on the simulated database as well as on the experimental database to determine the best wavelengths and, hence, the best performances in chlorophyll estimation for each index types. These indices go from simple reflectance ratios to more sophisticated indices using reflectance first derivatives (using the Savitzky and Golay method). We also tested other nondestructive methods to obtain total chlorophyll concentration: SPAD (Minolta Camera, Osaka, Japan) and neural networks. The validity of the actual PROSPECT model is challenged by our results: Important discordances are found when the indices are calculated with PROSPECT compared to experimental data, especially for some indices and wavelengths. The discordance is even greater when the indices are determined with PROSPECT and applied on the experimental database. A new calibration of PROSPECT is therefore necessary for any study aiming at using simulated spectra to determine or to calibrate indices. The “peak jump” and the multiple-peak feature observed on the first derivative of the reflectances (e.g., in the Red-Edge Inflection Point [REIP] index) has been investigated. It was shown that chlorophyll absorption alone can explain this feature. The peak jump disqualifies the REIP to be a valuable chlorophyll index. A simple modified difference ratio gave the best results among all published indices (cross-validated RMSE=2.1 μg/cm² on the experimental database). After calibration on the experimental database, modified Simple Ratio (mSR) and modified Normalized Difference (mND) indices gave the best performances (RMSECV=1.8 μg/cm² on the experimental database). The new Double Difference (DD) index, although not the best on the experimental database (RMSECV=2.9 μg/cm²), has the best results on the larger simulated database (RMSE=3.7 μg/cm²) and is expected to give good results on larger experimental databases. The best reflectance-based indices give better performances than the current commercial nondestructive device SPAD (RMSECV=4.5 μg/cm²). In this leaf-level study, the best indices are very near from each other, so that complex methods are useless: REIP-like, neural networks, and derivative-based indices are not necessary and give worst results than simpler properly chosen indices. These conclusions will certainly be different for a canopy-level study, where the derivative-based indices may perform significantly better than the other ones.     

Fuzzy learning vector quantization for hyperspectral coastal vegetation classification

Artificial neural networks (ANNs) may be of significant value in extracting vegetation type information in complex vegetation mapping problems, particularly in coastal wetland environments. Unsupervised, self-organizing ANNs have not been employed as frequently as supervised ANNs for vegetation mapping tasks, and further remote sensing research involving fuzzy ANNs is also needed. In this research, the utility of a fuzzy unsupervised ANN, specifically a fuzzy learning vector quantization (FLVQ) ANN, was investigated in the context of hyperspectral AVIRIS image classification. One key feature of the neural approach is that unlike conventional hyperspectral data processing methods, endmembers for a given scene, which can be difficult to determine with confidence, are not required for neural analysis. The classification accuracy of FLVQ was comparable to a conventional supervised multi-layer perceptron, trained with backpropagation (MLP) (KHAT (Kˆ) accuracy: 82.82% and 84.66%, respectively; normalized accuracy: 74.60% and 75.85%, respectively), with no significant difference at the 95% confidence level. All neural algorithms in the experiment yielded significantly higher classification accuracies than the conventional endmember-based hyperspectral mapping method assessed (i.e., matched filtering, where Kˆ accuracy = 61.00% and normalized accuracy = 57.96%). FLVQ was also dramatically more computationally efficient than the baseline supervised and unsupervised ANN algorithms tested, including the MLP and the Kohonen self-organizing map (SOM), respectively. The 400-neuron FLVQ network required only 3.6% of the computation time used by the MLP network, and only 5.9% of the MLP time was used by the 588-neuron FLVQ network. In addition, the 400-neuron FLVQ used only 16.7% of the time used by the 400-neuron SOM for model development.     

基于分形信号的高光谱影像增强方法

目前常用的高光谱影像增强方法大多继承了多光谱影像的增强处理方法,这类方法没有充分利用光谱信息,而基于混合像元分解的图像增强方法存在端元的选取问题。基于影像的自相似特征,探索运用分形信号进行遥感影像增强的可能性。以3景Hyperion高光谱影像数据为基础,把基于地毯的方法进行修正后用于计算高光谱影像中每一像元的分形信号。结果表明,与原始高光谱影像相比,分形信号影像可以更好地突出地物特征,从而达到影像增强的目的,原始曲线形态特征、初始尺度的选择以及采样点数目对分形信号和分形特征尺度均有影响。     

Multiple shadow fractions in spectral mixture analysis of a cotton canopy

Shadows are being used more frequently to estimate plant canopy biophysical characteristics. Typically, a zero value is assumed or a threshold value is derived from histogram analysis of imagery to determine the shadow endmember (EM). Here, two distinct shadow EMs were measured in situ for use in spectral mixture analysis of a cotton canopy on five dates in 2003. The four EMs used in the analysis were: sunlit green leaf, sunlit dry soil, self-shadowed leaf, shadowed dry soil. This 4-EM model was compared to a 3-EM model where a zero-value shade EM was used for unmixing with the two sunlit EMs. Multiple endmember spectral mixture analysis (MESMA) was used to allow EM composition to vary across each scene. The analysis and EMs were applied to fine-scale hyperspectral image data collected in the wavelength range, 440 to 810 nm. Ground data collected included percent cover, height, SPAD (a measure of leaf greenness), and chlorophyll a concentration. The normalized difference vegetation index (NDVI) was also compared to the unmixing results. Regression analysis showed that NDVI was equal to the 4-EM model for estimation of percent cover (r² = 0.95, RMSE = 6.6) although the NDVI y-intercept was closer to zero. The 4-EM model was best for estimating height (r² = 0.79, RMSE = 0.07 m) and chlorophyll a concentration (r² = 0.46, RMSE = 7.0 μg/cm²). The 3-EM model and NDVI performed poorly when estimating chlorophyll a concentration. Inclusion of two distinct shadow EMs in the model improved relationships to crop biophysical parameters and was better than assuming one, zero-value shade EM. Since MESMA operates at the pixel level and allows variable EM assignment to each pixel, mapping the spatial variability of shadows and other variables of interest is possible, providing a powerful input to canopy and ecosystem models as well as precision farming.     

基于核最小噪声分离变换的高光谱遥感影像特征提取研究

高光谱遥感影像具有高维非线性的特点,线性特征提取方法容易造成信息丢失和失真。在最小噪声分离变换(MNF)线性特征提取算法的基础上,引入核方法,提出核最小噪声分离变换(KMNF)高光谱遥感影像非线性特征提取方法。KMNF通过核函数,将样本映射到高维特征空间,在特征空间中运算线性MNF,实现原始空间中的非线性KMNF算法。进行基于KMNF的高光谱影像特征提取实验,分析样本个数对KMNF特征提取的效果,发现样本数量对KMNF特征提取的结果影响很小,较少的样本数即可达到较多样本时特征提取的效果。对比KMNF与MNF特征提取的效果,分析它们降维的效率与保留的信息量,发现KMNF总体降维效率与MNF相当,且体现出高光谱图像的非线性特征;在KMNF和MNF特征提取的基础上,利用SVM进行高光谱图像分类,KMNF+SVM的分类精度优于MNF+SVM。     

基于逐像素递归处理的高光谱实时亚像元目标检测

亚像元目标检测是高光谱图像应用的关键技术。由于高光谱数据的高维度增加了存储空间和数据处理的复杂度,实时处理成为了目标检测面临的重要问题。自适应匹配滤波算法(AMF)是一种有效的亚像元目标检测算法。在基于Woodbury引理实现以逐像素排列格式传输和存储的高光谱数据协方差矩阵实时求逆的基础上,以AMF为高光谱图像亚像元目标检测算法,推导出了基于逐像素递归处理的高光谱图像实时AMF目标检测流程。通过仿真数据和真实高光谱图像实验证明,相比于非实时AMF,实时AMF只需少量的存储空间便可得到同样甚至更高的检测精度 。