Digitální knihovnaUPCE
 

Fakulta elektrotechniky a informatiky / Faculty of Electrical Engineering and Informatics

Stálý URI pro tuto komunituhttps://hdl.handle.net/10195/3847

Práce obhájené před rokem 2008 jsou uloženy pouze v kolekci Vysokoškolské kvalifikační práce

Procházet

Search Results

Nyní se zobrazuje 1 - 10 z 13
  • Článekpeer-reviewedpostprint (accepted)Otevřený přístup
    RGB images-driven recognition of grapevine varieties using a densely connected convolutional network
    (Oxford University Press, 2022) Škrabánek, Pavel; Doležel, Petr; Matousek, Radomil
    We present a pocket-size densely connected convolutional network (DenseNet) directed to classification of size-normalized colour images according to varieties of grapes captured in those images. We compare the DenseNet with three established small-size networks in terms of performance, inference time and model size. We propose a data augmentation that we use in training the networks. We train and evaluate the networks on in-field images. The trained networks distinguish between seven grapevine varieties and background, where four and three varieties, respectively, are of red and green grapes. Compared to the established networks, the DenseNet is characterized by near state-of-the-art performance, short inference time and minimal model size. All these aspects qualify the network for real-time, mobile and edge computing applications. The DenseNet opens possibilities for constructing affordable selective harvesters in accordance with agriculture 4.0.
  • Článekpeer-reviewedpublishedOtevřený přístup
    Centroid based person detection using pixelwise prediction of the position
    (Elsevier Science BV, 2022) Doležel, Petr; Škrabánek, Pavel; Štursa, Dominik; Zanon, Bruno Baruque; Adrian, Hector Cogollos; Kryda, Pavel
    Implementations of person detection in tracking and counting systems tend towards processing of orthogonally captured images on edge computing devices. The ellipse-like shape of heads in orthogonally captured images inspired us to predict head centroids to determine positions of persons in images. We predict the centroids using a fully convolutional network (FCN). We combine the FCN with simple image processing operations to ensure fast inference of the detector. We experiment with the size of the FCN output to further decrease the inference time. We compare the proposed centroid-based detector with bounding box-based detectors on head detection task in terms of the inference time and the detection performance. We propose a performance measure which allows quantitative comparison of the two detection approaches. For the training and evaluation of the detectors, we form original datasets of 8000 annotated images, which are characterized by high variability in terms of lighting conditions, background, image quality, and elevation profile of scenes. We propose an approach which allows simultaneous annotation of the images for both bounding box-based and centroid-based detection. The centroid-based detector shows the best detection performance while keeping edge computing standards.
  • Konferenční objektpeer-reviewedpostprintOtevřený přístup
    RGB Images Driven Recognition of Grapevine Varieties
    (Springer Nature Switzerland AG, 2020) Škrabánek, Pavel; Doležel, Petr; Matoušek, Radomil; Junek, Petr
    We present a grapevine variety recognition system based on a densely connected convolutional network. The proposed solution is aimed as a data processing part of an affordable sensor for selective harvesters. The system classifies size normalized RGB images according to varieties of grapes captured in the images. We train and evaluate the system on in-field images of ripe grapes captured without any artificial lighting, in a direction of sunshine likewise in the opposite direction. A dataset created for this purpose consists of 7200 images classified into 8 categories. The system distinguishes among seven grapevine varieties and background, where four and three varieties have red and green grapes, respectively. Its average per-class classification accuracy is at 98.10% and 97.47% for red and green grapes, respectively. The system also well differentiates grapes from background. Its overall average per-class accuracy is over 98%. The evaluation results show that conventional cameras in combination with the proposed system allow construction of affordable automatic selective harvesters.
  • Článekpeer-reviewedpublished versionOtevřený přístup
    Person Detection for an Orthogonally Placed Monocular Camera
    (Hindawi limited, 2020) Škrabánek, Pavel; Doležel, Petr; Němec, Zdeněk; Štursa, Dominik
    Counting of passengers entering and exiting means of transport is one of the basic functionalities of passenger flow monitoring systems. Exact numbers of passengers are important in areas such as public transport surveillance, passenger flow prediction, transport planning, and transport vehicle load monitoring. To allow mass utilization of passenger flow monitoring systems, their cost must be low. As the overall price is mainly given by prices of the used sensor and processing unit, we propose the utilization of a visible spectrum camera and data processing algorithms of low time complexity to ensure a low price of the final product. To guarantee the anonymity of passengers, we suggest orthogonal scanning of a scene. As the precision of the counting is relevantly influenced by the precision of passenger recognition, we focus on the development of an appropriate recognition method. We present two opposite approaches which can be used for the passenger recognition in means of transport with and without entrance steps, or with split level flooring. The first approach is the utilization of an appropriate convolutional neural network (ConvNet), which is currently the prevailing approach in computer vision. The second approach is the utilization of histograms of oriented gradients (HOG) features in combination with a support vector machine classifier. This approach is a representative of classical methods. We study both approaches in terms of practical applications, where real-time processing of data is one of the basic assumptions. Specifically, we examine classification performance and time complexity of the approaches for various topologies and settings, respectively. For this purpose, we form and make publicly available a large-scale, class-balanced dataset of labelled RGB images. We demonstrate that, compared to ConvNets, the HOG-based passenger recognition is more suitable for practical applications. For an appropriate setting, it defeats the ConvNets in terms of time complexity while keeping excellent classification performance. To allow verification of theoretical findings, we construct an engineering prototype of the system.
  • Konferenční objektpeer-reviewedpostprintOtevřený přístup
    On possibilities of human head detection for person flow monitoring system
    (Springer Nature Switzerland AG, 2019) Doležel, Petr; Štursa, Dominik; Škrabánek, Pavel
    Along with the development of human society, economy, industry and engineering, as well as with growing population in the world's biggest cities, various approaches to person detection have become the subject of great interest. One approach to developing a person detection system is proposed in this paper. A high-angle video sequence is considered as the input to the system. Then, three classification algorithms are considered: support vector machines, pattern recognition neural networks and convolutional neural networks. The results showed very little difference between the classifiers, with the overall accuracy more than 95% over a testing set.
  • Článekpeer-reviewedpostprintOmezený přístup
    On Reporting performance of binary classifiers
    (2017) Škrabánek, Pavel; Doležel, Petr
    In this contribution, the question of reporting performance of binary classifiers is opened in context of the so called class imbalance problem. The class imbalance problem arises when a dataset with a highly imbalanced class distribution is used within the training or evaluation process. In such cases, only measures, which are not biased by distribution of classes in datasets, should be used; however, they cannot be chosen arbitrarily. They should be selected so that their outcomes provide desired information; and simultaneously, they should allow a full comparison of just evaluated classifier performance along, with performances of other solutions. As is shown in this article, the dilemma with reporting performance of binary classifiers can be solved using so called class balanced measures. The class balanced measures are generally applicable means, appropriate for reporting performance of binary classifiers on balanced as well as on imbalanced datasets. On the basis of the presented pieces of information, a suggestion for a generally applicable, fully-valued, reporting of binary classifiers performance is given.
  • Článekpeer-reviewedpublishedOtevřený přístup
    On reporting performance of binary classifiers
    (Univerzita Pardubice, 2017) Škrabánek, Pavel; Doležel, Petr
    In this contribution, the question of reporting performance of binary classifiers is opened in context of the so called class imbalance problem. The class imbalance problem arises when a dataset with a highly imbalanced class distribution is used within the training or evaluation process. In such cases, only measures, which are not biased by distribution of classes in datasets, should be used; however, they cannot be chosen arbitrarily. They should be selected so that their outcomes provide desired information; and simultaneously, they should allow a full comparison of just evaluated classifier performance along, with performances of other solutions. As is shown in this article, the dilemma with reporting performance of binary classifiers can be solved using so called class balanced measures. The class balanced measures are generally applicable means, appropriate for reporting performance of binary classifiers on balanced as well as on imbalanced datasets. On the basis of the presented pieces of information, a suggestion for a generally applicable, fully-valued, reporting of binary classifiers performance is given.
  • Článekpeer-reviewedpublishedOtevřený přístup
    Robust Grape Detector Based on SVMs and HOG Features
    (2017) Škrabánek, Pavel; Doležel, Petr
    Detection of grapes in real-life images is a serious task solved by researchers dealing with precision viticulture. In the case of white wine varieties, grape detectors based on SVMs classifiers, in combination with a HOG descriptor, have proven to be very efficient. Simplified versions of the detectors seem to be the best solution for practical applications. They offer the best known performance vs. time-complexity ratio. As our research showed, a conversion of RGB images to grayscale format, which is implemented at an image pre-processing level, is ideal means for further improvement of performance of the detectors. In order to enhance the ratio, we explored relevance of the conversion in a context of a detector potential sensitivity to a rotation of berries. For this purpose, we proposed a modification of the conversion, and we designed an appropriate method for a tuning of such modified detectors. To evaluate the effect of the new parameter space on their performance, we developed a specialized visualization method. In order to provide accurate results, we formed new datasets both for tuning and evaluation of the detectors. Our effort resulted in a robust grape detector which is less sensitive to image distortion.
  • Článekpeer-reviewedpostprintOtevřený přístup
    Weight Initialization Possibilities for Feedforward Neural Network with Linear Saturated Activation Functions
    (Elsevier Science BV, 2016) Doležel, Petr; Škrabánek, Pavel; Gago, Lumír
    Initial weight choice is an important aspect of the training mechanism for feedforward neural networks. This paper deals with a particular topology of a feedforward neural network, where symmetric linear saturated activation functions are used in a hidden layer. Training of such a topology is a tricky procedure, since the activation functions are not fully differentiable. Thus, a proper initialization method for that case is even more important, than dealing with neural networks with sigmoid activation functions. Therefore, several initialization possibilities are examined and tested here. As a result, particular initialization methods are recommended for application, according to the class of the task to be solved.
  • Konferenční objektpeer-reviewedpostprintOmezený přístup
    Pattern Recognition Neural Network as a Tool for Pest Birds Detection
    (IEEE (Institute of Electrical and Electronics Engineers), 2016) Doležel, Petr; Škrabánek, Pavel; Gago, Lumír
    Various kinds of vermin have been considered as a huge problem since primeval times. Over this period, means of protection against vermin have developed to be very quick and efficient. However, new goals in protection have appeared recently which reflects legislative changes in most countries. Public opinion has shifted towards greater environment protection. Nowadays, vermin control systems have turned from being used globally into local applications and from being applied preventively into casual usage. Thus, accurate vermin detection units are becoming very important parts of vermin control systems. This situation is valid in agricultural areas (e.g. vineyards) which are protected against pest birds, too. Reflecting on the current situation, a feedforward multilayer artificial neural network, aimed on detection of European starling in vineyards, is presented in this paper. Except a description and validation of the detection method, the idea of the comprehensive protection system is also outlined in this paper.