Detta arbete presenterar en icke-destruktiv metod för att detektera och klassificera mognadsgraden hos äpplen med användning av hyperspektrala bilder. Fastställning av mognadsgraden hos äpplen är intressant för bland annat äppelodlare och musterier vid lagring och beredning. Äpplens mognadsgrad är även intressant inom växtförädling. För att fastställa mognadsgraden idag krävs att det skärs i frukten, en så kallad destruktiv metod. Hyperspektrala bilder kan idag användas inom områden som jordbruk, miljöövervakning och militär spaning. / <p>Examensarbetet är utfört vid Institutionen för teknik och naturvetenskap (ITN) vid Tekniska fakulteten, Linköpings universitet</p>

In industry, it is crucial to be able to detect damage or abnormal behavior in machines. A machine's downtime can be minimized by detecting and repairing faulty components of the machine as early as possible. It is, however, economically inefficient and labor-intensive to detect machine fault sounds manual. In comparison with manual machine failure detection, automatic failure detection systems can reduce operating and personnel costs.  Although prior research has identified many methods to detect failures in drill machines using vibration or sound signals, this field still remains many challenges. Most previous research using machine learning techniques has been based on features that are extracted manually from the raw sound signals and classified using conventional classifiers (SVM, Gaussian mixture model, etc.). However, manual extraction and selection of features may be tedious for researchers, and their choices may be biased because it is difficult to identify which features are good and contain an essential description of sounds for classification. Recent studies have used LSTM, end-to-end 1D CNN, and 2D CNN as classifiers for classification, but these have limited accuracy for machine failure detection. Besides, machine failure occurs very rarely in the data. Moreover, the sounds in the real-world dataset have complex waveforms and usually are a combination of noise and sound presented at the same time. Given that drill failure detection is essential to apply in the industry to detect failures in machines, I felt compelled to propose a system that can detect anomalies in the drill machine effectively, especially for a small dataset. This thesis proposed modern artificial intelligence methods for the detection of drill failures using drill sounds provided by Valmet AB. Instead of using raw sound signals, the image representations of sound signals (Mel spectrograms and log-Mel spectrograms) were used as the input of my proposed models. For feature extraction, I proposed using deep learning 2-D convolutional neural networks (2D-CNN) to extract features from image representations of sound signals. To classify three classes in the dataset from Valmet AB (anomalous sounds, normal sounds, and irrelevant sounds), I proposed either using conventional machine learning classifiers (KNN, SVM, and linear discriminant) or a recurrent neural network (long short-term memory). For using conventional machine learning methods as classifiers, pre-trained VGG19 was used to extract features and neighborhood component analysis (NCA) as the feature selection. For using long short-term memory (LSTM), a small 2D-CNN was proposed to extract features and used an attention layer after LSTM to focus on the anomaly of the sound when the drill changes from normal to the broken state. Thus, my findings will allow readers to detect anomalies in drill machines better and develop a more cost-effective system that can be conducted well on a small dataset. There is always background noise and acoustic noise in sounds, which affect the accuracy of the classification system. My hypothesis was that noise suppression methods would improve the sound classification application's accuracy. The result of my research is a sound separation method using short-time Fourier transform (STFT) frames with overlapped content. Unlike traditional STFT conversion, in which every sound is converted into one image, a different approach is taken. In contrast, splitting the signal into many STFT frames can improve the accuracy of model prediction by increasing the variability of the data. Images of these frames separated into clean and noisy ones are saved as images, and subsequently fed into a pre-trained CNN for classification. This enables the classifier to become robust to noise. The FSDNoisy18k dataset is chosen in order to demonstrate the efficiency of the proposed method. In experiments using the proposed approach, 94.14 percent of 21 classes were classified successfully, including 20 classes of sound events and a noisy class. / <p>Vid tidpunkten för disputationen var följande delarbeten opublicerade: delarbete 2 och 3 inskickat.</p><p>At the time of the doctoral defence the following papers were unpublished: paper 2 and 3 submitted.</p> / AISound – Akustisk sensoruppsättning för AI-övervakningssystem / MiLo — miljön i kontrolloopen

Goal of this Diploma thesis was Convolutional neural network investigation in last years. Diploma thesis also contains information about designing of appropriate Convolutional neural network models and implementation of these models in Java programming language. Result of the thesis are comparison and evaluation of results which were reached from implemented application.

The master's thesis deals with detecting and tracking people in the video. To get optimal recognition was used convolution neural network, which extracts vector features from the enclosed frame the face. The extracted vector is further classified. Recognition process must take place in a real time and also with respect are selected optimal methods. There is a new dataset faces, which was obtained from a video record at the faculty area. Videos and dataset were used for experiments to verify the accuracy of the created system. The recognition accuracy is about 85% . The proposed system can be used, for example, to register people, counting passages or to report the occurrence of an unknown person in a building.

This master's thesis deals with design and implementation of convolutional neural networks used in person re-identification. Implemented convolutional neural networks were tested on two datasets CUHK01 a CUHK03. Results, comparable with state of the art methods were acheved on these datasets. Designed networks were implemented in Caffe framework.

The Swedish armed forces use the Single Source Intelligent Cell (SSIC), developed by Saab, for analysis of aerial reconnaissance video and report generation. The analysis can be time-consuming and demanding for a human operator. In the analysis workflow, identifying vehicles is an important part of the work. Artificial Intelligence is widely used for analysis in many industries to aid or replace a human worker. In this paper, the possibility to aid the human operator with air reconnaissance data analysis is investigated, specifically, object detection for finding cars in aerial images. Many state-of-the-art object detection models for vehicle detection in aerial images are based on a Convolutional Neural Network (CNN) architecture. The Faster R-CNN- and SSD-based models are both based on this architecture and are implemented. Comprehensive experiments are conducted using the models on two different datasets, the open Video Verification of Identity (VIVID) dataset and a confidential dataset provided by Saab. The datasets are similar, both consisting of aerial images with vehicles. The initial experiments are conducted to find suitable configurations for the proposed models. Finally, an experiment is conducted to compare the performance of a human operator and a machine. The results from this work prove that object detection can be used to supporting the work of air reconnaissance image analysis regarding inference time. The current performance of the object detectors makes applications, where speed is more important than accuracy, most suitable.

In a known-item search task (KIS), the goal is to find a previously seen image in a multimedia collection. In this thesis, we discuss two different approaches based on the visual description of the image. In the first one, the user creates a collage of images (using images from an external search engine), based on which we provide the most similar results from the dataset. Our results show that preprocessing the images in the dataset by splitting them into several parts is a better way to work with the spatial information contained in the user input. We compared the approach to a baseline, which does not utilize this spatial information and an approach that alters a layer in a deep neural network. We also present an alternative approach to the KIS task, search by faces. In this approach, we work with the faces extracted from the images. We investigate face representation for the ability to sort the faces based on their similarity. Then we present a structure that allows easy exploration of the set of faces. We provide a demo, implementing all presented techniques.

This study has investigated whether it is possible to classify time series data originating from a gamified learning management system. By using the school data provided by the gamification company Insert Coin AB, the aim was to distribute the teacher’s supervision more efficiently among students who are more likely to fail. Motivating this is the possibility that the student retention and completion rate can be increased. This was done by using Long short-term memory and convolutional neural networks and Markov chain to classify time series of event data. Since the classes are balanced the classification was evaluated using only the accuracy metric. The results for the neural networks show positive results but overfitting seems to occur strongly for the convolutional network and less so for the Long short-term memory network. The Markov chain show potential but further work is needed to mitigate the problem of a strong correlation between sequence length and likelihood.

Machine learning and image recognition is a big and growing subject in today's society. Therefore the aim of this thesis is to compare convolutional neural networks with different hyperparameter settings and see how the hyperparameters affect the networks test accuracy in identifying images of traffic signs. The reason why traffic signs are chosen as objects to evaluate hyperparameters is due to the author's previous experience in the domain. The object itself that is used for image recognition does not matter. Any dataset with images can be used to see the hyperparameters affect. Grid search is used to create a large amount of models with different width and depth, learning rate and momentum. Convolution layers, activation functions and batch size are all tested separately. These experiments make it possible to evaluate how the hyperparameters affect the networks in their performance of recognizing images of traffic signs. The models are created using Keras API and then trained and tested on the dataset Traffic Signs Preprocessed. The results show that hyperparameters affect test accuracy, some affect more than others. Configuring learning rate and momentum can in some cases result in disastrous results if they are set too high or too low. Activation function also show to be a crucial hyperparameter where it in some cases produce terrible results.

Three-dimensional information is often used in robotics and 3D-mapping. There exist several ways to obtain a three-dimensional map. However, the time of flight used in the laser scanners or the structured light utilized by Kinect-like sensors sometimes are not sufficient. In this thesis, we investigate two CNN based stereo matching methods for obtaining 3D-information from a grayscaled pair of rectified images.While the state-of-the-art stereo matching method utilize a Siamese architecture, in this project a two-channel and a two stream network are trained in an attempt to outperform the state-of-the-art. A set of experiments were performed to achieve optimal hyperparameters. By changing one parameter at the time, the networks with architectures mentioned above are trained. After a completed training the networks are evaluated with two criteria, the error rate, and the runtime.Due to time limitations, we were not able to find optimal learning parameters. However, by using settings from [17] we train a two-channel network that performed almost on the same level as the state-of-the-art. The error rate on the test data for our best architecture is 2.64% while the error rate for the state-of-the-art Siamese network is 2.62%. We were not able to achieve better performance than the state-of-the-art, but we believe that it is possible to reduce the error rate further. On the other hand, the state-of-the-art Siamese stereo matching network is more efficient and faster during the disparity estimation. Therefore, if the time efficiency is prioritized, the Siamese based network should be considered.