Grzegorz Wieczorek

Over the past few years, significant investments in smart traffic monitoring systems have been made. The most important step in machine learning is detecting and recognizing objects relative to vehicles. Due to variations in vision and different lighting conditions, the recognition and tracking of vehicles under varying extreme conditions has become one of the most challenging tasks. To deal with this, our proposed system presents an adaptive method for robustly recognizing several existing automobiles in dense traffic settings. Additionally, this research presents a broad framework for effective on-road vehicle recognition and detection. Furthermore, the proposed system focuses on challenges typically noticed in analyzing traffic scenes captured by in-vehicle cameras, such as consistent extraction of features. First, we performed frame conversion, background subtraction, and object shape optimization as preprocessing steps. Next, two important features (energy and deep optical flow) were extracted. The incorporation of energy and dense optical flow features in distance-adaptive window areas and subsequent processing over the fused features resulted in a greater capacity for discrimination. Next, a graph-mining-based approach was applied to select optimal features. Finally, the artificial neural network was adopted for detection and classification. The experimental results show significant performance in two benchmark datasets, including the LISA and KITTI 7 databases. The LISA dataset achieved a mean recognition rate of 93.75% on the LDB1 and LDB2 databases, whereas KITTI attained 82.85% accuracy on separate training of ANN.

In this article, an approach to drill wear evaluation is presented. Tool condition monitoring is an important problem in furniture manufacturing and similar industries. At the same time, approaches that rely on sets of sensors, often tend to be to robust or complex for the production environment. Instead of signals acquired from dedicated sensors, presented approach uses images of drilled holes as input data. Initial pictures are processed and enhanced in order to highlight the crucial properties. A set of selected features is then calculated on the resulting images, and later used during the training of 5 state-of-the-art classifiers. Presented research also evaluates number of images for consecutive drillings that needs to be taken into account in order to produce accurate results. From the selected set, the best performing classifier was Random Forest and it achieved close to 100% accuracy.

In this paper, a novel approach to evaluation of feature extraction methodologies is presented. In the case of machine learning algorithms, extracting and using the most efficient features is one of the key problems that can significantly influence overall performance. It is especially the case with parameter-heavy problems, such as tool condition monitoring. In the presented case, images of drilled holes are considered, where state of the edge and the overall size of imperfections have high influence on product quality. Finding and using a set of features that accurately describes the differences between the edge that is acceptable or too damaged is not always straightforward. The presented approach focuses on detailed evaluation of various feature extraction approaches. Each chosen method produced a set of features, which was then used to train a selected set of classifiers. Five initial feature sets were obtained, and additional ones were derived from them. Different voting methods were used for ensemble approaches. In total, 38 versions of the classifiers were created and evaluated. Best accuracy was obtained by the ensemble approach based on Weighted Voting methodology. A significant difference was shown between different feature extraction methods, with a total difference of 11.14% between the worst and best feature set, as well as a further 0.2% improvement achieved by using the best voting approach.

The multiclass prediction approach to the problem of recognizing the state of the drill by classifying images of drilled holes into three classes is presented. Expert judgement was made on the basis of the quality of the hole, by dividing the collected photographs into the classes: “very fine,” “acceptable,” and “unacceptable.” The aim of the research was to create a model capable of identifying different levels of quality of the holes, where the reduced quality would serve as a warning that the drill is about to wear down. This could reduce the damage caused by a blunt tool. To perform this task, real-world data were gathered, normalized, and scaled down, and additional instances were created with the use of data-augmentation techniques, a self-developed transformation, and with general adversarial networks. This approach also allowed us to achieve a slight rebalance of the dataset, by creating higher numbers of images belonging to the less-represented classes. The datasets generated were then fed into a series of convolutional neural networks, with different numbers of convolution layers used, modelled to carry out the multiclass prediction. The performance of the so-designed model was compared to predictions generated by Microsoft’s Custom Vision service, trained on the same data, which was treated as the benchmark. Several trained models obtained by adjusting the structure and hyperparameters of the model were able to provide better recognition of less-represented classes than the benchmark.

The problem of segmenting the cross-section through the longissimus muscle in beef carcasses with computer vision methods was investigated. The available data were 111 images of cross-sections coming from 28 cows (typically four images per cow). Training data were the pixels of the muscles, marked manually. The AlexNet deep convolutional neural network was used as the classifier, and single pixels were the classified objects. Each pixel was presented to the network together with its small circular neighbourhood, and with its context represented by the further neighbourhood, darkened by halving the image intensity. The average classification accuracy was 96%. The accuracy without darkening the context was found to be smaller, with a small but statistically significant difference. The segmentation of the longissimus muscle is the introductory stage for the next steps of assessing the quality of beef for the alimentary purposes.

In this paper we present a new segmentation method meant for boost area that remains after removing the tumour using BCT (breast conserving therapy). The selected area is a region on which radiation treatment will later be made. Consequently, an inaccurate designation of this region can result in a treatment missing its target or focusing on healthy breast tissue that otherwise could be spared. Needless to say that exact indication of boost area is an extremely important aspect of the entire medical procedure, where a better definition can lead to optimizing of the coverage of the target volume and, in result, can save normal breast tissue. Precise definition of this area has a potential to both improve the local control of the disease and to ensure better cosmetic outcome for the patient. In our approach we use U-net along with Keras and TensorFlow systems to tailor a precise solution for the indication of the boost area. During the training process we utilize a set of CT images, where each of them came with a contour assigned by an expert. We wanted to achieve a segmentation result as close to given contour as possible. With a rather small initial data set we used data augmentation techniques to increase the number of training examples, while the final outcomes were evaluated according to their similarity to the ones produced by experts, by calculating the mean square error and the structural similarity index (SSIM).

In this paper we introduce the enhanced drill wear recognition method, based on classifiers ensemble, obtained using transfer learning and data augmentation methods. Red, green and yellow classes are used to describe the current drill state. The first one corresponds to the case when drill should be immediately replaced. The second one denotes a tool that is still in a good condition. The final class refers to the case when a drill is suspected of being worn out, and a human expert evaluation would be required. The proposed algorithm uses three different, pretrained network models and adjusts them to the drill wear classification problem. To ensure satisfactory results, each of the methods used was required to achieve accuracy above 90% for the given classification task. Final evaluation is achieved by voting of all three classifiers. Since the initial data set was small (242 instances), the data augmentation method was used to artificially increase the total number of drill hole images. The experiments performed confirmed that the presented approach can achieve high accuracy, even with such a limited set of training data.

This paper presents an improved method for recognizing the drill state on the basis of hole images drilled in a laminated chipboard, using convolutional neural network (CNN) and data augmentation techniques. Three classes were used to describe the drill state: red - for drill that is worn out and should be replaced, yellow - for state in which the system should send a warning to the operator, indicating that this element should be checked manually, and green - denoting the drill that is still in good condition, which allows for further use in the production process. The presented method combines the advantages of transfer learning and data augmentation methods to improve the accuracy of the received evaluations. In contrast to the classical deep learning methods, transfer learning requires much smaller training data sets to achieve acceptable results. At the same time, data augmentation customized for drill wear recognition makes it possible to expand the original dataset and to improve the overall accuracy. The experiments performed have confirmed the suitability of the presented approach to accurate class recognition in the given problem, even while using a small original dataset.

Textural features based upon thresholding and run length encoding have been successfully applied to the problem of classification of the quality of lacquered surfaces in furniture exhibiting the surface defect known as orange skin. The set of features for one surface patch consists of 12 real numbers. The classifier used was the one nearest neighbour classifier without feature selection. The classification quality was tested on 808 images 300 by 300 pixels, made under controlled, close-to-tangential lighting, with three classes: good, acceptable and bad, in close to balanced numbers. The classification accuracy was not smaller than 98% when the tested surface was not rotated with respect to the training samples, 97% for rotations up to 20 degrees and 95.5% in the worst case for arbitrary rotations.