No description available.

No description available.

With the advancement of autonomous driving research, 3D detection based on LiDAR point cloud has gradually become one of the top research topics in the field of artificial intelligence. Compared with RGB cameras, LiDAR point cloud can provide depth information, while RGB images can provide denser resolution. Features from LiDAR and cameras are considered to be complementary. However, due to the sparsity of the LiDAR point clouds, a dense and accurate RGB/3D projective relationship is difficult to establish especially for distant scene points. Recent works try to solve this problem by designing a network that learns missing points or dense point density distribution to compensate for the sparsity of the LiDAR point cloud. During the master’s exploration, we consider addressing this problem from two aspects. The first is to design a GAN(Generative Adversarial Network)-based module to reconstruct point clouds, and the second is to apply regional point cloud enhancement based on motion maps. In the first aspect, we propose to use an imagine-and-locate process, called UYI. The objective of this module is to improve the point cloud quality and is independent of the detection stage used for inference. We accomplish this task through a GAN-based cross-modality module that uses image as input to infer a dense LiDAR shape. In another aspect, inspired by the attention mechanism of human eyes, we use motion maps to perform random augmentation on point clouds in a targeted manner named motion map-assisted enhancement, MAE. Boosted by our UYI and MAE module, our experiments show a significant performance improvement in all tested baseline models. In fact, benefiting from the plug-and-play characteristics of our module, we were able to push the performance of the existing state-of-the-art model to a new height. Our method not only has made great progress in the detection performance of vehicle objects but also achieved an even bigger leap forward in the pedestrian category. In future research, we will continue to explore the feasibility of spatio-temporal correlation methods in 3D detection, and 3D detection related to motion information extraction could be a promising direction.

Artificial neural network (ANN) is a machine learning approach where parameters, i.e., frequency parameters and amplitude parameters, are learnt during the training process. Random features model is a special case of ANN that the structure of random features model is as same as ANN’s but the parameters’ learning processes are different. For random features model, the amplitude parameters are learnt during the training process but the frequency parameters are sampled from some distributions. If the frequency distribution of the random features model is well-chosen, both models can approximate data well. Adaptive random Fourier features with Metropolis sampling is an enhanced random Fourier features model which can select appropriate frequency distribution adaptively. This thesis studies Rectified Linear Unit and sigmoid features and combines them with the adaptive idea to generate another two adaptive random features models. The results show that using the particular set of hyper-parameters, adaptive random Rectified Linear Unit features model can also approximate the data relatively well, though the adaptive random Fourier features model performs slightly better. / I artificiella neurala nätverk (ANN), som används inom maskininlärning, behöver parametrar, kallade frekvensparametrar och amplitudparametrar, hittasgenom en så kallad träningsprocess. Random feature-modeller är ett specialfall av ANN där träningen sker på ett annat sätt. I dessa modeller tränasamplitudparametrarna medan frekvensparametrarna samplas från någon sannolikhetstäthet. Om denna sannolikhetstäthet valts med omsorg kommer båda träningsmodellerna att ge god approximation av givna data. Metoden Adaptiv random Fourier feature[1] uppdaterar frekvensfördelningen adaptivt. Denna uppsats studerar aktiveringsfunktionerna ReLU och sigmoid och kombinerar dem med den adaptiva iden i [1] för att generera två ytterligare Random feature-modeller. Resultaten visar att om samma hyperparametrar som i [1] används så kan den adaptiva ReLU features-modellen approximera data relativt väl, även om Fourier features-modellen ger något bättre resultat.

Every year there are thousands of aviation accidents along with hundreds of human deaths that happen around the world. While the data is sparse, it is well documented that many of these happen from emergency landings gone awry. While pilots can generally make great landings in clear daytime conditions, they are significantly handicapped when it comes to landing at night or amongst poor visibility conditions. Due to the nature of this problem and some of the large scale advances in software technology we propose a solution that provides a significant improvement from the status quo. Using transfer learning on neural networks to classify satellite images along with terrain elevation data from the USGS we are able to recreate maps that can readily direct pilots to locations that are relatively flat and lack structures or vegetation. Using San Luis Obispo as our data set we confirmed that we could correctly classify at least 93\% of landable terrain and then identified areas within that area that could safely be used to land a plane. We then transfer this data into a 3D rendering program that allows us to visualize what is happening. In additional to visualizing where the landing paths are we also create a landing algorithm that demonstrates how the plane will maintain its current glide path and navigate to a successful landing while avoiding obstacles.

No description available.

No description available.

<p>Roller bearings are the common components used in the mechanical systems for mechanical processing and production. The running state of roller bearings often determines the machining accuracy and productivity on a manufacturing line. Roller bearing failure may lead to the shutdown of production lines, resulting in serious economic losses. Therefore, the research on roller bearing fault diagnosis has a great value. This thesis research first proposes a method of signal frequency spectral resampling to tackle the problem of bearing fault detection at different rotating speeds using a single speed dataset for training the network such as the one dimensional convolutional neural network (1D CNN). Second, this research work proposes a technique to connect the graph structures constructed from spectral components of the different bearing fault frequency bands into a sparse graph structure, so that the fault identification can be carried out effectively through a graph neural network in terms of the computation load and classification rate. Finally, the frequency spectral resampling method for feature extraction is validated using our self-collected datasets. The performance of the graph neural network with our proposed sparse graph structure is validated using the Case Western Reserve University (CWRU) dataset as well as our self-collected datasets. The results show that our proposed method achieves higher bearing fault classification accuracy than those recently proposed by other researchers using machine learning approaches and neural networks.</p>

With the inception of music streaming and media content delivery platforms, there has been a tremendous increase in the music available on the internet and the metadata associated with it. In this study, we address the problem of violin artist identification, which tries to classify the performing artist based on the learned features. Even though numerous previous works studied the problem in detail and developed features and deep learning models that can be used, an interesting fact was that most studies focused on artist identification in western popular music and less on Indian classical music. For the same reason, there was no standardized dataset for this purpose. Hence, we curated a new dataset consisting of audio recordings from 6 renowned South Indian Carnatic violin artists. In this study, we explore the use of log-Mel-spectrogram feature and the embeddings generated by a pre-learned VGGish network on a Convolutional Neural Network and Convolutional Recurrent Neural Network Model. From the experiments, we observe that the Convolutional Recurrent Neural Network model trained using the log-Mel-spectrogram feature gave the optimal performance with a classification accuracy of 71.70%. / Med starten av plattformar för musikströmning och leverans av mediainnehåll har det skett en enorm ökning av musiken tillgänglig på internet och den metadata som är associerad med den. I denna studie tar vi upp problemet med fiolkonstnärsidentifikation, som försöker klassificera den utövande konstnären utifrån de inlärda dragen. Även om många tidigare verk studerade problemet i detalj och utvecklade funktioner och modeller för djupinlärning som kan användas, var ett intressant faktum att de flesta studier fokuserade på artistidentifiering i västerländsk populärmusik och mindre på indisk klassisk musik. Av samma anledning fanns det ingen standardiserad datauppsättning för detta ändamål. Därför kurerade vi en ny datauppsättning bestående av ljudinspelningar från 6 kända sydindiska karnatiska violinkonstnärer. I den här studien utforskar vi användningen av log-Melspektrogramfunktionen och inbäddningarna som genereras av ett förinlärt VGGishnätverk på ett Convolutional Neural Network och Convolutional Recurrent Neural Network Model. Från experimenten observerar vi att modellen Convolutional Recurrent Neural Network tränad med hjälp av log-Mel-spektrogramfunktionen gav optimal prestanda med en klassificeringsnoggrannhet på 71,70%.

Record linkage, in a genealogical context, is the process of identifying individuals from multiple sources which refer to the same real-world entity. Current solutions focus on the individuals in question and on complex rules developed by human experts. Genealogical databases are highly-structured with relationships existing between the individuals and other instances. These relationships can be utilized and human involvement greatly minimized by using a filtered structured neural network. These neural networks, using traditional back-propagation methods, are biased in a way to make the network human readable. The results show an increase in precision and recall when pedigree data is available and used.