Shigeki Nakauchi

Near infrared spectroscopy and hyperspectral imaging for prediction and visualisation of fat and fatty acid content in intact raw beef cuts

The meat quality grade of a beef carcass is greatly affected by its visible fat content. In premium beef from Japanese Black (Wagyu) cattle, a high fat content is greatly valued. However, the fatty acid composition, which is linked to the properties of the fat, is not considered in grading. In this paper, we describe the feasibility of an evaluation method based on food composition and its distribution. An intact raw beef cut from Wagyu cattle was used as an evaluation target. A total of 90 samples from various parts of three Wagyu cattle were measured by near infrared (NIR) hyperspectral imaging at wavelengths of 1000–2300 nm at a spatial resolution of 380 urn pixel−1 and were also analysed by conventional physical and chemical methods. The fat and fatty acid content were selected as the objective content, including the proportions of total saturated fatty acid (SFA), total unsaturated fatty acid (UFA) and the main fatty acids: myristic [C14:0, where Cx:y indicates the number of carbon atoms (x) and the number of double bonds (y)], palmitic (C16:0), stearic (C18:0), myristoleic (C14:1), palmitoleic (C16:1), oleic (C18:1) and linoleic (C18:2). The mean spectrum from an area extracted from the hyperspectral image to fit the area analysed by physical and chemical methods was used to develop partial least squares regression models for prediction of fat and fatty acid content. The prediction of total fat, SFA and UFA were satisfactory with r2, standard error of prediction (SEP) and ratio of prediction to deviation (RPD) values of 0.90, 0.87 and 0.89, 4.81%, 1.69% and 3.41% and 2.84, 2.43 and 2.84, respectively. For individual fatty acids, the r2 and RPD values ranged from 0.68 to 0.89 and 1.69 to 2.85, respectively. Prediction of fat content for each pixel of the hyperspectral image made using these prediction models yielded spatially distributed visualisations of the content. These results showed the feasibility of a beef evaluation method based on fat content evaluated by NIR hyperspectral imaging.

Reconstruction of Munsell color space by a five-layer neural network

We have constructed a wine-glass-type five-layer neural network and generated an identity mapping of the surface spectral-reflectance data of 1280 Munsell color chips, using a backpropagation learning algorithm. To achieve an identity mapping, the same data set is used for the input and for the teacher. After the learning was completed, we analyzed the responses to individual chips of the three hidden units in the middle layer in order to obtain the internal representation of the color information. We found that each of the three hidden units corresponds to a psychological color attribute, that is, the Munsell value (luminance), red–green, and yellow–blue. We also examined the relationship between the internal representation and the number of hidden units and found that the network with three hidden units acquires optimum color representation. The five-layer neural network is shown to be an efficient method for reproducing the transformation of color information (or color coding) in the visual system.

Computational theory of color transparency: recovery of spectral properties for overlapping surfaces

A computational theory of color transparency for color images containing X junctions is described. This theory is based on physical models of color transparency under conditions of additive or subtractive color mixture that describe the relationship among four colors at an X junction. Algorithms are derived for recovering transmittance and surface reflectance functions of a transparent medium from a set of sensor responses at an X junction. The algorithms are based on the ability to describe surface reflectance and transmittance functions by using a linear combination of orthogonal basis functions. We also address algorithms for determination of depth ordering of overlapping surfaces and the type of color mixture by checking the physical realizability of recovered functions.

Reconstruction of Munsell color space by a five-layered neural network

A wine-glass-type five-layered neural network (81-10-3-10-81) has been constructed, and identity mapping has been realized on the set of surface spectral reflectance data of Munsell color chips by a backpropagation learning algorithm. The network is divided into two parts: encoder (81-10-3) and decoder (3-10-81). Surface spectral reflectance data as physical attributes of color are transformed nonlinearly in each part. After identity mapping learning was completed, the response pattern of the three hidden units in the middle layer was analyzed to obtain the internal representation of color information acquired by self-learning. As a result, it was found that each hidden unit responds to psychological color attributes, that is, one for the value and the other two units for the constant value plane of the Munsell color system which consists of the hue and chroma. The nonlinear analysis method using five-layered neural networks is shown to be an efficient method for elucidating the color information coding mechanisms in the visual system

Selection of optimal combinations of band-pass filters for ice detection by hyperspectral imaging

Hyperspectral imaging captures rich information in spatial and spectral domains but involves high costs and complex data processing. The use of a set of optical band-pass filters (BPFs) in the acquisition of spectral images is proposed for reducing dimensionality of spectral data while maintaining target detection and/or categorization performance. A set of BPFs that could distinguish ice from surrounding water on various materials (e.g., asphalt), was designed as an example. Relatively high accuracy (90.28%) was achieved with only two BPFs, showing the potential of this approach for accurate target detection with lesser complexity than conventional methods.

Optimization of excitation–emission band-pass filter for visualization of viable bacteria distribution on the surface of pork meat

A novel method of optically reducing the dimensionality of an excitation-emission matrix (EEM) by optimizing the excitation and emission band-pass filters was proposed and applied to the visualization of viable bacteria on pork. Filters were designed theoretically using an EEM data set for evaluating colony-forming units on pork samples assuming signal-to-noise ratios of 100, 316, or 1000. These filters were evaluated using newly measured EEM images. The filters designed for S/N = 100 performed the best and allowed the visualization of viable bacteria distributions. The proposed method is expected to be a breakthrough in the application of EEM imaging.

Effects of color information on face processing using event-related potentials and gamma oscillations.

In humans, face configuration, contour and color may affect face perception, which is important for social interactions. This study aimed to determine the effect of color information on face perception by measuring event-related potentials (ERPs) during the presentation of natural- and bluish-colored faces. Our results demonstrated that the amplitude of the N170 event-related potential, which correlates strongly with face processing, was higher in response to a bluish-colored face than to a natural-colored face. However, gamma-band activity was insensitive to the deviation from a natural face color. These results indicated that color information affects the N170 associated with a face detection mechanism, which suggests that face color is important for face detection.

Acquisition of color opponent representation by a three-layered neural network model

This paper discusses color representation in the visual system by analysis of a three-layered neural network model. The model incorporates physiological knowledge of color representation at the sensor level (broad-band trichromatic representation by cones) and the higher level (narrow-band color representation by color-coded cells in V4). We trained the model to perform a mapping between these color representations by the back propagation algorithm and analyzed the acquired characteristics of the hidden units. It turned out that the hidden units learned characteristics similar to those of the color opponent cells found in the visual system. It was concluded that the R-G and Y-B color opponent representations reflect the efficiency of the color representation in the visual system from investigations on the efficiency of color representation in the hidden layer and on the capability of the color recognition task of the model.

The face-selective N170 component is modulated by facial color

Faces play an important role in social interaction by conveying information and emotion. Of the various components of the face, color particularly provides important clues with regard to perception of age, sex, health status, and attractiveness. In event-related potential (ERP) studies, the N170 component has been identified as face-selective. To determine the effect of color on face processing, we investigated the modulation of N170 by facial color. We recorded ERPs while subjects viewed facial color stimuli at 8 hue angles, which were generated by rotating the original facial color distribution around the white point by 45° for each human face. Responses to facial color were localized to the left, but not to the right hemisphere. N170 amplitudes gradually increased in proportion to the increase in hue angle from the natural-colored face. This suggests that N170 amplitude in the left hemisphere reflects processing of facial color information.

Temporal properties of material categorization and material rating: visual vs non-visual material features

Humans can visually recognize material categories of objects, such as glass, stone, and plastic, easily. However, little is known about the kinds of surface quality features that contribute to such material class recognition. In this paper, we examine the relationship between perceptual surface features and material category discrimination performance for pictures of materials, focusing on temporal aspects, including reaction time and effects of stimulus duration. The stimuli were pictures of objects with an identical shape but made of different materials that could be categorized into seven classes (glass, plastic, metal, stone, wood, leather, and fabric). In a pre-experiment, observers rated the pictures on nine surface features, including visual (e.g., glossiness and transparency) and non-visual features (e.g., heaviness and warmness), on a 7-point scale. In the main experiments, observers judged whether two simultaneously presented pictures were classified as the same or different material category. Reaction times and effects of stimulus duration were measured. The results showed that visual feature ratings were correlated with material discrimination performance for short reaction times or short stimulus durations, while non-visual feature ratings were correlated only with performance for long reaction times or long stimulus durations. These results suggest that the mechanisms underlying visual and non-visual feature processing may differ in terms of processing time, although the cause is unclear. Visual surface features may mainly contribute to material recognition in daily life, while non-visual features may contribute only weakly, if at all.