1 - 10 of 15
-
PublicationPhenoliner 2.0: RGB and near-infrared (NIR) image acquisition for an efficient phenotyping in grapevine research( 2021)
-
PublicationDetection of pyrrolizidine alkaloid containing herbs using hyperspectral imaging in the short-wave infrared( 2021)Plants containing pyrrolizidine alkaloids (PA) are unwanted contaminants in consumer products such as herbal tea due to their toxicity to humans. The detection of these plants or their components using hyperspectral imaging was investigated, with focus on application in sensor-based sorting. For this, 431hyperspectral images of leafs from three common herbs (peppermint, lemon balm, stinging nettle) and the poisonous common groundsel were acquired. By using a convolutional neural network, a mean F1 score of 0.89 was obtained for the classification of all four plant products based on the individual spectra. To validate the neural network, significant wavelengths were determined and visualized in an attribution map.
-
PublicationDeveloping a handheld NIR sensor for the detection of ripening in grapevine( 2021)It has already been proven that near infrared (NIR) reflectance spectroscopy can be used to measure the ripeness of grapes by the determination of reducing sugar and acid contents. Until now, winegrowers need to collect a random one hundred berries sample per plot, to measure these parameters destructively for the estimation of the ideal harvest time of the gained product. Meanwhile, inexpensive sensors are available, to build convenient instruments for the non-destructive, low-priced and fast control of ripening parameters in the vineyard. For this, a small device including a NIR sensor (900 nm - 1700 nm / 1300 nm - 2600 nm) was built from a Raspberry Pi 3 and a NIR sensor. Spectra of individual berries, sampled from six different Vitis vinifera (L.) cultivars (Riesling, Chardonnay, Pinot Noir, Dornfelder, Pinot Gris and Dakapo) were collected. Corresponding reference data were determined with high performance liquid chromatography (HPLC). Samples were taken from different fruit-, as well as cluster zones and from the beginning of veraison until after harvest, to ensure a broad range of ingredients and the ripening properties of different berries from the vine. White, as well as red varieties were used to establish the built sensor as a viable tool for ripening prediction for mainly cultivated vines. Spectra of teinturier berries with strongly coloured flesh or skin were collected to verify its accuracy for these cultivars, too. This study is the first that systematically investigates the ripening parameters of a whole vineyard with a handheld sensor. The sensor can be used in viticulture practice to detect the ripening process and ideal harvest time due to effectiveness and simplicity.
-
PublicationAutomatic visual inspection based on trajectory data( 2019)Automatic inspection tasks have successfully been implemented in several industrial fields and are of growing importance. Visual inspection using optical sensors is wide spread due to the vast variety of different sensors, observable features and comparatively low prices. It seems obvious that corresponding systems are blind towards mechanical features and inspection of those typically requires highly specialized, inflexible and costly systems. Recently, we have shown in the context of sensor-based sorting that tracking objects over a time period allows deriving motion-based features which potentially enable discrimination of optically identical objects, although an optical sensor is used. In this paper, we take one step back from the specific application and study the classification of test objects based on their trajectories. The objects are observed while receiving a certain impulse. We further refrain from manually designing features but use raw coordinates as extracted from a series of images. The success of the method is demonstrated by discriminating spheres made of similar plastic types while bouncing off a plane.
-
PublicationApplication of area-scan sensors in sensor-based sorting( 2018)In the field of machine vision, sensor-based sorting is an important real-time application that enables the separation of a material feed into different classes. While state-of-the-art systems utilize scanning sensors such as line-scan cameras, advances in sensor technology have made application of area scanning sensors feasible. Provided a sufficiently high frame rate, objects can be observed at multiple points in time. By applying multiobject tracking, information about the objects contained in the material stream can be fused over time. Based on this information, our approach further allows predicting the position of each object for future points in time. While conventional systems typically apply a global, rather simple motion model, our approach includes an individual motion model for each object, which in turn allows estimating the point in time as well as the position when reaching the separation stage. In this contribution, we present results from our collaborative research project and summarize the present advances by discussing the potential of the application of area-scan sensors for sensor-based sorting. Among others, we introduce our simulation-driven approach and present results for physical separation efficiency for simulation-generated data, demonstrate the potential of using motion-based features for material classification and discuss real-time related challenges.
-
PublicationImproving material characterization in sensor-based sorting by utilizing motion information( 2017)Sensor-based sorting provides state-of-the-art solutions for sorting of cohesive, granular materials. Systems are tailored to a task at hand, for instance by means of sensors and implementation of data analysis. Conventional systems utilize scanning sensors which do not allow for extraction of motion related information of objects contained in a material feed. Recently, usage of area-scan cameras to overcome this disadvantage has been proposed. Multitarget tracking can then be used in order to accurately estimate the point in time and position at which any object will reach the separation stage. In this paper, utilizing motion information of objects which can be retrieved from multitarget tracking for the purpose of classification is proposed. Results show that corresponding features can significantly increase classification performance and eventually decrease the detection error of a sorting system.
-
PublicationNumerical investigation of optical sorting using the discrete element method( 2017)Automated optical sorting systems are important devices in the growing field of bulk solids handling. The initial sorter calibration and the precise optical sorting of many materials is still very time consuming and difficult. A numerical model of an automated optical belt sorter is presented in this study. The sorter and particle interaction is described with the Discrete Element Method (DEM) while the separation phase is considered in a post processing step. Different operating parameters and their influence on sorting quality are investigated. In addition, two models for detecting and predicting the particle movement between the detection point and the separation step are presented and compared, namely a conventional line scan camera model and a new approach combining an area scan camera model with particle tracking.
-
PublicationImproving multitarget tracking using orientation estimates for sorting bulk materials( 2017)Optical belt sorters can be used to sort a large variety of bulk materials. By the use of sophisticated algorithms, the performance of the complex machinery can be further improved. Recently, we have proposed an extension to industrial optical belt sorters that involves tracking the individual particles on the belt using an area scan camera. If the estimated behavior of the particles matches the true behavior, the reliability of the separation process can be improved. The approach relies on multitarget tracking using hard association decisions between the tracks and the measurements. In this paper, we propose to include the orientation in the assessment of the compatibility of a track and a measurement. This allows us to achieve more reliable associations, facilitating a higher accuracy of the tracking results.
-
PublicationSimulation of an inverse schlieren image acquisition system for inspecting transparent objects( 2016)This paper describes a novel approach for inspecting transparent objects. Since the underlying optical setup is based on a schlieren setup, any defect can be detected that leads to the deflection or extinction of incident light rays. By capturing a light field image of a defect-free transparent object and by illuminating objects under test with that very light field, the proposed inspection system can visualize defects by acquiring a single inspection image only. In order to evaluate the presented approach, a physically based rendering framework is used. It is extended by models and implementations of the necessary emitter and sensor plugins. Simulation experiments with virtual scenes that consist of a double-convex lens affected by different types of defects are the basis for a qualitative evaluation of the proposed method. The results show that a single image acquired with the described method is sufficient to test transparent objects for defects that cause the deflection or extinction of rays, e.g., enclosed absorbing or scattering impurities, shape anomalies, differences of the index of refraction and 3D-misalignments.
-
PublicationFeature selection with a budget( 2016)Feature selection is an important step in all practical applications of pattern recognition. As such, it is not surprising that during the past decades it has received a lot of attention from the research community. The topic is well understood and many methods have been put to the test. Most methods, however, overlook an aspect critical to real-time applications: limited computation time. The set of selected features must not only be suitable to solve the task, but must also ensure that the task can be solved within the available time. With this in mind, we propose a method for feature selection with a budget. We approach the problem by stating feature selection as a multi-objective optimization problem. This problem is solved using the well known NSGA-II algorithm. We evaluate our approach using one synthetic and two real-world datasets. We explore the properties of the genetic algorithm and investigate the classification performance of the resulting selections. Our results show that the selected feature sets are highly suitable, especially when considering real-time systems.
