More than 20 years ago when Mr. K. H. Norris firstly introduced the near infrared spectroscopy (NIRS) as a powerful technology in the field of composition analysis of cereals those who were interested in the area of classical spectroscopy would not like to recognize its potential. This tendency still remains at present however it leaves no room for doubt that from viewpoints of applied spectroscopy the NIRS has consolidated its position. From a viewpoint of NIRS application in the field of nondestructive or non invasive measuring techniques history of this technology is only the last decade in Japan. However since the technology was firstly introduced to composition analysis of agricultural commodities in the same manner as in other countries R and D have been growing more actively in diversified fields such as agriculture and industry as well as medical science. In addition the NIRS technology are becoming of general interest by combining other techniques to create various hyphenated instrumentations such as FTNIR MCFTNIR NIRCT and NIR-NMR. In this paper new trends of R D on NIR spectroscopy which are being conducted in Japan will be reviewed. 2. S1JMMARY OF PRESENT R D ON NIRS IN JAPAN NIRS applications reported in the last 3 years are summarized in Table 1. Table 1 Applications of NIRS in Japan Application for Agriculture Taste evaluation of rice and coffee Determination of chemical compositions rice for breeding Determination of chemical compositions in tea Determination of sugar contents in intact peaches Japanese pears Satsuma oranges and apples Determination of sugars and acids in intact tomatoes Determination of forage composition Application for Industry Analysis of state of water in foods Application of analyzing Maillard Reaction''s Process Pattern recognition of NIR spectra as related to process control of roasting coffee beans Quality control of tea processing Determination of moisture content of Surimi products 2 / SPIE Vol. 1379 Optics in Agriculture (1990)
A high intensity computer controlled spectrophotometer was used to obtain near infrared (700 to 1 100 nm) speciral transmittance data of intact samples of onions, cantaloupes and potatoes. The dry matter was determined by freeze drying and the soluble solids by refractometer. A data reduction program computed the ratio of two second derivatives, opümizing 5 parameters of the computation. Linear correlations were above 0.99 for onions.
Inages from conventional video systems are being digitized in coraputers for the analysis of small trash particles in cotton. The method has been developed to automate particle counting and area measurements for bales of cotton prepared for market. Because the video output is linearly proportional to the amount of light reflected the best spectral band for optimum particle discrimination should be centered at the wavelength of maximum difference between particles and their surroundings. However due to the spectral distribution of the illumination energy and the detector sensitivity peak image performance bands were altered. Reflectance from seven mechanically cleaned cotton lint samples and trash removed were examined for spectral contrast in the wavelength range of camera sensitivity. Pixel intensity histograms from the video systent are reported for simulated trashmeter area reference samples (painted dots on panels) and for cotton containing trash to demonstrate the particle discrimination mechanism. 2.
Two studies were performed to investigate the feasibility of using near infrared reflectance spectroscopy (NIRS) with undried silages. In the first study silages were analyzed for major components (e. g. dry matter crude protein and other forms of nitrogen fiber and in vitro digestible dry matter) and short chain fatty acids (SCFA). NIRS was found to operate satisfactorily except for some forms of nitrogen and SCFA. In study two various methods of grinding spectral regions and sample presentation were examined. Undried Wiley ground samples in a rectangular cell gave the best overall results for non-dry ice undried grinds with wavelengths between 1100 and 2498 nm. Silages scanned after drying however produced the best results. Intact samples did not perform as well as ground samples and wavelengths below 1100 nm were of little use. 2 .
Recently near infrared (NTR) reflectance instrumentation has been used to provide an empirical measure of wheat hardness. This hardness scale is based on the radiation scattering properties of meal particles at 1680 and 2230 nm. Hard wheats have a larger mean particles size (PS) after grinding than soft wheats. However wheat kernel moisture content can influence mean PS after grinding. The objective of this study was to determine the sensitivity of MR wheat hardness measurements to moisture content and to make the hardness score independent of moisture by correcting hardness measurements for the actual moisture content of measured samples. Forty wheat cultivars composed of hard red winter hard red spring soft red winter and soft white winter were used. Wheat kernel subsamples were stored at 20 40 60 and 80 relative humidity (RH). After equilibration samples were ground and the meal analyzed for hardness score (HS) and moisture. HS were 48 50 54 and 65 for 20 40 60 and 80 RH respectively. Differences in HS within each wheat class were the result of a moisture induced change in the PS of the meal. An algorithm was developed to correct HS to 11 moisture. This correction provides HS that are nearly independent of moisture content. 1.
Over the past 10 years near-infrared researchers predominantly have been concerned with mathematical manipulations which provide more reliable calibrations. Considerable emphasis in the last five year has been placed on the use of principal components (PCA) and partial least squares (PLS) both being artificial variates computed from Log (1/R) data as a viable manipulation for solving some of the problems associated with qualitative and quantitative NIR analyses. However it remains to be proven that the variates from PCA and PLS are better for quantitative analysis than the original Log (1/R) data from which the variates are computed. Fourier analysis is proposed in this paper as an alternative data treatment for both quantitative and qualitative work. Spectra of whole-kernel and ground wheat with protein analyses are used to demonstrate the use of Fourier transforms in the analysis of NIR data. 1.
As the sizes of today''s farms increase larger equipment is required to ensure the timeliness of field operations. The operation of this equipment requires a significant amount of concentration by the operator. As a result automatic guidance of agricultural equipment is a very promising method of increasing the farmer''s productivity. The objective of this research is to develop and evaluate real-time image processing techniques to extract guidance information from digitized video images of tilled and untilled soil and standing crop and stubble1. The guidance information consists of the vehicle''s heading and lateral offset errors. These errors can be used to supply the operator with a visual indication of the amount of overlap or missing occurring. Ultimately automatic steering of the vehicle will be possible. Still video images of tilled and untilled soil and standing crop and stubble were gathered in the field then digitized and stored on computer disk. The images were then analyzed using three image processing algorithms two of which were developed by the authors. The results indicate that the heading errors as predicted by the algorithms agreed with the visually estimated heading errors to within 2 on average. Agreement of 0. 1 m or less (on average) was observed between the offset errors as predicted by the algorithms and the visually estimated offset errors. The computation times of each algorithm do not currently meet real-time requirements however for
Dennis St. George John Feddes (Dept. of Agricultural Engineering University of Alberta Edmonton AB Canada T6G 2Hl) A prototype light collection and transmission device was developed and evaluated for the potential of irradiating plants grown in an opague growth chamber. Results indicated that the device transmitted light with a photon flux of 130 1amol/s/m2 (4000-7000 nm) to the bottom of the growth chamber when direct solar radiation was 800 W/m2 (300-2500 nm) outside. The overall collection and transmission efficiency for photosynthetically active radiation is 19. 2. A growth trial with plants indicated that artificial lighting is required during cloudy periods. 1.
Diffuse thickness (the sample thickness necessary to bring about complete diffusion of a directed incident light beam) was used to predict the distribution of transmitted radiation in a light scattering medium. The results show that for nonabsorbing wavelengths the shape of the sample is a major factor in determining the distribution of radiation in the medium. As absorption increases the distribution becomes less a function of sample shape. 1.
The effects of removing illumination gradients on the spatial distribution characteristics of visible fat within beef longissimus dorsi muscle were investigated. Gradients were assumed to be slowly varying and were removed by subtracting a smoothed version of the image from the original. Two types of filters were used to smooth the images: a linear lowpass filter and a greyscale morphological opening. The effects of the two filters on the binary images after thresholding using a Boolean random set model were investigated. Results showed that the choice of filter significantly influenced model parameter estimates. It was concluded that the morphological filter provided the most consistent means of eliminating adverse effects of illumination gradients. 1.
Various vision methods for inspecting the growth and quality of poinsettia plants are discussed in this paper . The visible and near-infrared vision approaches are based on previous spectral reflectance measurements . Low (0 ppm) nitrogen plants grown in a greenhouse showed an increase in red (0. 7 - 0. 75 rim) and a decrease in near-infrared ( 0 . 8 - 1 . 1 im) reflectance over high ( 256 ppm) nitrogen levels . Growth chamber plants showed similar reflectance in the red but different NIR reflectance than with greenhouse plants . NIR reflectance was affected by vegetative density and not by leaf nitrogen content. Thermal imaging techniques (12 - 14 im) improve canopy temperature measurements . The usefulness of image methods depends on reflectivity analog-digital sensitivity and background lighting quality. An electronic plant doctor based on a database of images would be a useful tool for the grower to perform visual diagnostics. 1.
Ground and aerial experiments were conducted with color (NC) color infrared (CIR) and black and white film and video systems to compare the limitations! advantages of each method of image acquisition with photographs of natural vegetation including cypress stands wetlands and cultivated crops such as: tomatoes cucumbers and citrus. Image analysis with a Linear Measuring System (LMS) and a scanning densitometer were used to quantify healthy stressed and diseased foliage!canopy of each crop for comparisons with visual estimates. videography and photography were useful in delineating topographic features and location of vegetation. The NC video systems yielded images that distinctly separated healthy and dying foliage but did not compare with the CIR video or photography in outlining distinct areas of stress and disease. Aerial photography provided a synoptic view of the fields and cypress stands not otherwise possible. CIR images were easier to process with the LMS than NC video or photographic frames. CIR video and photographic systems produced clearer differences between healthy and stressed foliage. Spectral curves produced with the scanning densitometer correlated well with visual grading of health and stress. . 2.
Active mid-infrared backscatter characteristics of various agricultural crop leaves were measured at different wavelengths in the 9-11 m spectral range at varied angles of incidence under both co-polarized and cross-polarized conditions. Measurements indicate that differences exist in the backscatter signatures between crop species and also among different strains of the same species. Backscatter is also seen to depend on the leaf moisture. Our preliminary studies demonstrate the potential of using active mid-infrared backscatter characteristics to remotely identify certain crop fypes and monitor their leaf moisture content. 1.
Methods for quantifying watercore damage in Red Delicious apples are described that measures visible and/or near-infrared (NIR) radiation transmitted through the core of individual apples. A linear regression of the data for NIR light alone and on the data for the combination of NIR and visible light produced R squared values from . 90 to . 96 indicating little difference between the different data sets. The combination of visible and NIR works equally as well as NIR alone. Images captured by a low-light level and an image-intensifier equipped black and white camera are compared to study the effect of sensor intensity sensitivity in the classification process. The intensified image results show the mean probability of an error as 1 3. 2 percent for under classifying the five different classes of watercore damage compared to 24. 9 and 27. 5 (NIR NIR and visible) for the low - light level camera. The mean probability of an error of over classifying from the intensified images is 17. 1 percent compared to 24. 0 and 27. 2 (NIR NIR and visible) for the low -light level camera. Increases in intensity sensitivity can improve the classification of watercore damage. Two computational methods to numerically quantify watercore damage are examined and compared. The first method finds the mean grey level within a 130 pixel (rows) by 100 pixel (columns) window centered about the stem end of
Peanut kernels from grade samples were sorted into damaged and undamaged categories based on their optical characteristics. A machine vision system used grey level information to detect certain types of damage. Also color coordinate information collected from a colorimeter provided additional damage information. Certain damage categories were correctly classified with 95 accuracy. 1 .
A machine vision system was developed to inspect fresh market carrots. It was designed to grade carrots with an axial and transverse resolution of 0. 5mmper pixel. Hardware consisted of camera digital signal processing (DSP) imaging board host computer and illumination components. Feature extraction methods detect the major defects. A Bayes classification technique was used to construct the decision function which classify carrots as acceptable or cull. The system was able to image and classify in approximately 2. 5carrots/second. 1.
Morphometrical features of single grain kernels or particles were used to discriminate two visibly similar wheat varieties foreign material in wheat hardsoft and spring-winter wheat classes and whole from broken corn kernels. Milled fractions of hard and soft wheat were evaluated using textural image analysis. Color image analysis of sound and mold damaged corn kernels yielded high recognition rates. The studies collectively demonstrate the potential for automated classification and assessment of grain quality using image analysis.
The acousto-optic tunable filter (AOTF) was demonstrated to be an effective and economical means for implementing a compact ( field deployable ) near-infrared Acousto-Optical Spectrometer (AOS) for quality assessment of small fruits. The design and construction of this device including design limitations and problems are presented. The characterization of the AOTF and the AOS system design address the areas of radiation source selection optical alignment optical design signal detection computer control and system testing. This work confirmed the operating specifications and limitations of the AOTF as described by the manufacturer.
Australia grows large quantities of radiata pine for domestic consumption and a significant proportion of this is graded to an Australian Standard Appearance Grade. This paper describes automating the visual inspection of this timber in order to speed processing and improve quality control of the product. The requirement is to detect and identify the visual features on the surface of the timber after the surface has been dressed. These features include sound and encased knots of various sizes pith bark bluestain holes and wane. The image is captured using a linear array CCD camera as the board moves underneath on a conveyor belt. The first stage is to detect the areas that contain features. The image is divided into smaller local areas and first and second order statistical measures are calculated. These form the input to a neural network that has been trained to classify the local areas into clear and feature areas. The choice of measures is crucial to the ability of the neural network to perform the classification of local areas. The second stage is to determine the type of feature in the feature local areas. Various methods are employed to determine a threshold that segments the feature correctly. The size of the feature can be used to identify it uniquely. The list of features and their positions forms the input to the grading program. The grading rules defmed
Radiograms of individual wheat kernels some of which contained immature insects (weevils moths or borers) were captured using microscopy into a 512x512x8 frame buffer 16. 4 pm/pixel. The resulting image was reduced repeatedly to half-size to produce collages containing 4 16 64 and 256 reduced images. Half the collages contained one and half no insect. Subjects were shown sets of 500 images on a 13 " monitor and were asked to determine whether an insect was present. Since hidden insects are typically characterized by a light insect inside a dark ring (the hole) inside a light kernel we measured the size of the recognizable region as the length (major diameter) and width of the hole in pixels in the reduced image. Contrast was represented as the square root of the variance of the pixel intensity along length or width. We find that recognition can be predicted to 80-90 by length alone contrast or width add little. The length for 50 recognition amounts to 12. 20. 6 22. 10. 7 and 19. 80. 6 pixels for weevils moths and borers resp. It is suggested that insect recognition depends on recognizing as yet unknown image features which are lost when images are reduced beyond this size. 1."
The rate at which pigs expose themselves to their thermal environment provides a parameter of interest in thermoregulation-. related research. A set of algorithms was developed to compensate for the distortion introduced under test conditions in the relevant image information. Restoration was obtained by reconstruction into three dimensional space of the scene based on knowledge of both camera behaviour and " pig-edge shape " . 1. PROBLEM AND OBJECTIVE Boon''s experimental data (figure la) suggest that the floor area covered by a group of resting pigs in a standardized test set-up provides a quantitative indication of their thermal need at least as accurate as his more intricately vision-related ''huddling index'' concept1. Bruce and Boon2 interprete the observed behaviour as the establishment of a situation of minimal discomfort involving a balance between a " mechanical discomfort component " - linked to group crowding when reducing exposure to thermal environment - and a remaining " thermal discomfort component " . Figure lb depicts a transposition of the curves deduced from the experimental data in figure la (supplementedwith extra knowledge on the pigs'' thermobehaviour) onto a reference scale more purely reflecting the involved external demands: - Mechanical discomfort as approximately reflected in group weight divided by projected group area (given weight this parameter is extractable from image data). - Thermal demand estimated based on the position of the ambient temperature relative to the temperature
This paper reports on the development of a model-based technique to locate pigs in fairly unstructured scenes. Model-based image processing is potentially a very powerful technique for identifying and classifying images. It is particularly relevant for biological objects such as animals which are difficult to define in numerical terms. Here the model is based on an image of a typical pig viewed from above . The model can then be rotated translated scaled and bent laterally to find a good match within an image of another pig. The output of the model helps to segment and classify the pig and the information can be used to guide further localized image processing. The development of the model and methods of fitting it to the image are described and results on a sample image are presented. 1.
Many components of animal production have been automated. For example weighing feeding identification and yield recording on cattle pigs poultry and fish. However some of these tasks still require a considerable degree of human input and more effective automation could lead to better husbandry. For example if the weight of pigs could be monitored more often without increasing labour input then this information could be used to measure growth rates and control fat level allowing accurate prediction of market dates and optimum carcass quality to be achieved with improved welfare at minimum cost. Some aspects of animal production have defied automation. For example attending to the well being of housed animals is the preserve of the expert stockman. He gathers visual data about the animals in his charge (in more plain words goes and looks at their condition and behaviour) and processes this data to draw conclusions and take actions. Automatically collecting data on well being implies that the animals are not disturbed from their normal environment otherwise false conclusions will be drawn. Computer image analysis could provide the data required without the need to disturb the animals. This paper describes new work at the Institute of Engineering Research which uses image analysis to estimate the weight of pigs as a starting point for the wider range of applications which have been identified. In particular a technique has been developed to
The reflectance ratio meter, an optical device for detecting weeds by measuring the ratio of reflected red and near-infrared light, is described. Experiments were conducted to evaluate the accuracy of the sensor in detecting weeds in the inter-row of growing soybeans and to characterize its sensitivity. Several methods for interpreting the signal were evaluated. The reflectance ratio meter is shown to have potential for estimating local weed populations.
The feasibility investigation of using machine vision technology to locate corn plants is an important issue for field production automation in the agricultural industry. This paper presents an approach which was developed to locate the center of a corn plant using image processing techniques. Corn plants were first identified using a main vein detection algorithm by detecting a local feature of corn leaves leaf main veins based on the spectral difference between mains and leaves then the center of the plant could be located using a center locating algorithm by tracing and extending each detected vein line and evaluating the center of the plant from intersection points of those lines. The experimental results show the usefulness of the algorithm for machine vision applications related to corn plant identification. Such a technique can be used for pre. cisc spraying of pesticides or biotech chemicals. 1.
A machine vision system was designed. for determining the peanut pod ripeness color after the outer hull layer of the pod had been removed. Pods were fed from a vibratory bowl feeder through a singulation chute to a measurement chamber where they entered free-fall. An IR detector sensed the presence of the peanut pod leaving the singulation chute and triggered a B/W CCD camera to capture the image. A ring strobe surrounding the camera lens illuminated the peanut. Algorithms were developed to determine the pod ripeness color for three views of the pod. 1.