Taste Recognition in E-Tongue Using Local Discriminant Preservation Projection

Mastication-Enhanced Taste-Based Classification of Multi-Ingredient Dishes for Robotic Cooking

Chefs frequently rely on their taste to assess the content and flavor of dishes during cooking. While tasting the food, the mastication process also provides continuous feedback by exposing the taste receptors to food at various stages of chewing. Since different ingredients of the dish undergo specific changes during chewing, the mastication helps to understand the food content. The current methods of electronic tasting, on the contrary, always use a single taste snapshot of a homogenized sample. We propose a robotic setup that uses the mixing to imitate mastication and tastes the dish at two different mastication phases. Each tasting is done using a conductance probe measuring conductance at multiple, spatially distributed points. This data is used to classify 9 varieties of scrambled eggs with tomatoes. We test four different tasting methods and analyze the resulting classification performance, showing a significant improvement over tasting homogenized samples. The experimental results show that tasting at two states of mechanical processing of the food increased classification F1 score to 0.93 in comparison to the traditional tasting of a homogenized sample resulting in F1 score of 0.55. We attribute this performance increase to the fact that different dishes are affected differently by the mixing process, and have different spatial distributions of the salinity. It helps the robot to distinguish between dishes of the same average salinity, but different content of ingredients. This work demonstrates that mastication plays an important role in robotic tasting and implementing it can improve the tasting ability of robotic chefs.

Digital Taste in Mulsemedia Augmented Reality: Perspective on Developments and Challenges

Digitalization of human taste has been on the back burners of multi-sensory media until the beginning of the decade, with audio, video, and haptic input/output(I/O) taking over as the major sensory mechanisms. This article reviews the consolidated literature on augmented reality (AR) in the modulation and stimulation of the sensation of taste in humans using low-amplitude electrical signals. Describing multiple factors that combine to produce a single taste, various techniques to stimulate/modulate taste artificially are described. The article explores techniques from prominent research pools with an inclination towards taste modulation. The goal is to seamlessly integrate gustatory augmentation into the commercial market. It highlights core benefits and limitations and proposes feasible extensions to the already established technological architecture for taste stimulation and modulation, namely, from the Internet of Things, artificial intelligence, and machine learning. Past research on taste has had a more software-oriented approach, with a few trends getting exceptions presented as taste modulation hardware. Using modern technological extensions, the medium of taste has the potential to merge with audio and video data streams as a viable multichannel medium for the transfer of sensory information.

Hierarchical Boosting Dual-Stage Feature Reduction Ensemble Model for Parkinson's Disease Speech Data

As a neurodegenerative disease, Parkinson's disease (PD) is hard to identify at the early stage, while using speech data to build a machine learning diagnosis model has proved effective in its early diagnosis. However, speech data show high degrees of redundancy, repetition, and unnecessary noise, which influence the accuracy of diagnosis results. Although feature reduction (FR) could alleviate this issue, the traditional FR is one-sided (traditional feature extraction could construct high-quality features without feature preference, while traditional feature selection could achieve feature preference but could not construct high-quality features). To address this issue, the Hierarchical Boosting Dual-Stage Feature Reduction Ensemble Model (HBD-SFREM) is proposed in this paper. The major contributions of HBD-SFREM are as follows: (1) The instance space of the deep hierarchy is built by an iterative deep extraction mechanism. (2) The manifold features extraction method embeds the nearest neighbor feature preference method to form the dual-stage feature reduction pair. (3) The dual-stage feature reduction pair is iteratively performed by the AdaBoost mechanism to obtain instances features with higher quality, thus achieving a substantial improvement in model recognition accuracy. (4) The deep hierarchy instance space is integrated into the original instance space to improve the generalization of the algorithm. Three PD speech datasets and a self-collected dataset are used to test HBD-SFREM in this paper. Compared with other FR algorithms and deep learning algorithms, the accuracy of HBD-SFREM in PD speech recognition is improved significantly and would not be affected by a small sample dataset. Thus, HBD-SFREM could give a reference for other related studies.

AdvKin: Adversarial Convolutional Network for Kinship Verification

Kinship verification in the wild is an interesting and challenging problem. The goal of kinship verification is to determine whether a pair of faces are blood relatives or not. Most previous methods for kinship verification can be divided as handcrafted features-based shallow learning methods and convolutional neural network (CNN)-based deep-learning methods. Nevertheless, these methods are still facing the challenging task of recognizing kinship cues from facial images. The reason is that the family ID information and the distribution difference of pairwise kin-faces are rarely considered in kinship verification tasks. To this end, a family ID-based adversarial convolutional network (AdvKin) method focused on discriminative Kin features is proposed for both small-scale and large-scale kinship verification in this article. The merits of this article are four-fold: 1) for kin-relation discovery, a simple yet effective self-adversarial mechanism based on a negative maximum mean discrepancy (NMMD) loss is formulated as attacks in the first fully connected layer; 2) a pairwise contrastive loss and family ID-based softmax loss are jointly formulated in the second and third fully connected layer, respectively, for supervised training; 3) a two-stream network architecture with residual connections is proposed in AdvKin; and 4) for more fine-grained deep kin-feature augmentation, an ensemble of patch-wise AdvKin networks is proposed (E-AdvKin). Extensive experiments on 4 small-scale benchmark KinFace datasets and 1 large-scale families in the wild (FIW) dataset from the first Large-Scale Kinship Recognition Data Challenge, show the superiority of our proposed AdvKin model over other state-of-the-art approaches.

Structural Damage Classification in a Jacket-Type Wind-Turbine Foundation Using Principal Component Analysis and Extreme Gradient Boosting

Damage classification is an important topic in the development of structural health monitoring systems. When applied to wind-turbine foundations, it provides information about the state of the structure, helps in maintenance, and prevents catastrophic failures. A data-driven pattern-recognition methodology for structural damage classification was developed in this study. The proposed methodology involves several stages: (1) data acquisition, (2) data arrangement, (3) data normalization through the mean-centered unitary group-scaling method, (4) linear feature extraction, (5) classification using the extreme gradient boosting machine learning classifier, and (6) validation applying a 5-fold cross-validation technique. The linear feature extraction capabilities of principal component analysis are employed; the original data of 58,008 features is reduced to only 21 features. The methodology is validated with an experimental test performed in a small-scale wind-turbine foundation structure that simulates the perturbation effects caused by wind and marine waves by applying an unknown white noise signal excitation to the structure. A vibration-response methodology is selected for collecting accelerometer data from both the healthy structure and the structure subjected to four different damage scenarios. The datasets are satisfactorily classified, with performance measures over 99.9% after using the proposed damage classification methodology.

Nonlinear Feature Extraction Through Manifold Learning in an Electronic Tongue Classification Task

A nonlinear feature extraction-based approach using manifold learning algorithms is developed in order to improve the classification accuracy in an electronic tongue sensor array. The developed signal processing methodology is composed of four stages: data unfolding, scaling, feature extraction, and classification. This study aims to compare seven manifold learning algorithms: Isomap, Laplacian Eigenmaps, Locally Linear Embedding (LLE), modified LLE, Hessian LLE, Local Tangent Space Alignment (LTSA), and t-Distributed Stochastic Neighbor Embedding (t-SNE) to find the best classification accuracy in a multifrequency large-amplitude pulse voltammetry electronic tongue. A sensitivity study of the parameters of each manifold learning algorithm is also included. A data set of seven different aqueous matrices is used to validate the proposed data processing methodology. A leave-one-out cross validation was employed in 63 samples. The best accuracy (96.83%) was obtained when the methodology uses Mean-Centered Group Scaling (MCGS) for data normalization, the t-SNE algorithm for feature extraction, and k-nearest neighbors (kNN) as classifier.

Electronic Tongue Recognition with Feature Specificity Enhancement

As a kind of intelligent instrument, an electronic tongue (E-tongue) realizes liquid analysis with an electrode-sensor array and certain machine learning methods. The large amplitude pulse voltammetry (LAPV) is a regular E-tongue type that prefers to collect a large amount of response data at a high sampling frequency within a short time. Therefore, a fast and effective feature extraction method is necessary for machine learning methods. Considering the fact that massive common-mode components (high correlated signals) in the sensor-array responses would depress the recognition performance of the machine learning models, we have proposed an alternative feature extraction method named feature specificity enhancement (FSE) for feature specificity enhancement and feature dimension reduction. The proposed FSE method highlights the specificity signals by eliminating the common mode signals on paired sensor responses. Meanwhile, the radial basis function is utilized to project the original features into a nonlinear space. Furthermore, we selected the kernel extreme learning machine (KELM) as the recognition part owing to its fast speed and excellent flexibility. Two datasets from LAPV E-tongues have been adopted for the evaluation of the machine-learning models. One is collected by a designed E-tongue for beverage identification and the other one is a public benchmark. For performance comparison, we introduced several machine-learning models consisting of different combinations of feature extraction and recognition methods. The experimental results show that the proposed FSE coupled with KELM demonstrates obvious superiority to other models in accuracy, time consumption and memory cost. Additionally, low parameter sensitivity of the proposed model has been demonstrated as well.

Fuzzy Evaluation Output of Taste Information for Liquor Using Electronic Tongue Based on Cloud Model

As a taste bionic system, electronic tongues can be used to derive taste information for different types of food. On this basis, we have carried forward the work by making it, in addition to the ability of accurately distinguish samples, be more expressive by speaking evaluative language like human beings. Thus, this paper demonstrates the correlation between the qualitative digital output of the taste bionic system and the fuzzy evaluation language that conform to the human perception mode. First, through principal component analysis (PCA), backward cloud generator and forward cloud generator, two-dimensional cloud droplet groups of different flavor information were established by using liquor taste data collected by electronic tongue. Second, the frequency and order of the evaluation words for different flavor of liquor were obtained by counting and analyzing the data appeared in the artificial sensory evaluation experiment. According to the frequency and order of words, the cloud droplet range corresponding to each word was calculated in the cloud drop group. Finally, the fuzzy evaluations that originated from the eight groups of liquor data with different flavor were compared with the artificial sense, and the results indicated that the model developed in this work is capable of outputting fuzzy evaluation that is consistent with human perception rather than digital output. To sum up, this method enabled the electronic tongue system to generate an output, which conforms to human's descriptive language, making food detection technology a step closer to human perception.

Optimally Distributed Kalman Filtering with Data-Driven Communication

For multisensor data fusion, distributed state estimation techniques that enable a local processing of sensor data are the means of choice in order to minimize storage and communication costs. In particular, a distributed implementation of the optimal Kalman filter has recently been developed. A significant disadvantage of this algorithm is that the fusion center needs access to each node so as to compute a consistent state estimate, which requires full communication each time an estimate is requested. In this article, different extensions of the optimally distributed Kalman filter are proposed that employ data-driven transmission schemes in order to reduce communication expenses. As a first relaxation of the full-rate communication scheme, it can be shown that each node only has to transmit every second time step without endangering consistency of the fusion result. Also, two data-driven algorithms are introduced that even allow for lower transmission rates, and bounds are derived to guarantee consistent fusion results. Simulations demonstrate that the data-driven distributed filtering schemes can outperform a centralized Kalman filter that requires each measurement to be sent to the center node.