Search

Image segmentation is one of the key components in systems performing computer vision recognition tasks. Various algorithms for image segmentation have been developed in the literature. Among them, more recently, deep learning algorithms have been remarkably successful in performing this task. A downside with deep neural networks for segmentation is that they require a large amount of labeled dataset for training. This prerequisite is one of the main reasons that led researchers to adopt data augmentation approaches in order to minimize manual labeling efforts while maintaining highly accurate results. This paper uses classical non-deep learning methods for background extraction to increase the size of the dataset used to train deep learning attention segmentation algorithms when images are presented as time-series to the model. The method presented adopts the Gaussian mixture-based (MOG2) foreground-background segmentation followed by dilation and erosion to create masks necessary to train the deep learning models. It is applied in the context of planktonic images captured in situ as time series. Various evaluation metrics and visual inspection are used to compare the performance of the deep learning algorithms. Experimental results show higher accuracy achieved by the deep learning algorithms for time-series image attention segmentation when the proposed data augmentation methodology is utilized to increase the training dataset.

Master's thesis in Automation and Signal Processing Bladder canceristhefourth most common type of cancer in men and the eighth in women. Patient treated for this cancer must be monitored for the rest of their live due to the recurrence of this disease.That need for monitoring makes bladder cancer the most expensive cancer. Pathologists analyse histological images of tissues to evaluate grade of the cancer.This analyse is done manually, therefore it depends on depathologist and it is not reproducible. To fix that problem, in this thesis an automatic and reproducible method is proposed. The method proposed in this thesis use classical image processing and deep neural networks to compute the number and type of cells in a histological image. First, using classical image processing cells from the input images are automatically labelled, then that labels are manually corrected to create a ground truth. Using that ground truth a Faster Convolutional Neural Network is trained and a model is generated. That model is trained and tested in different scenarios and results are compared to decide which one is the best. Classical image processing shows a precision of 96.6% and recall of 64.6% detecting non-positive cells, while for positive those values are 82.4% and 66.7%, respectively. Using the best neuralnet work model the precision and recall detecting non-positive cells was 4.5% and 71.7%, while for positive cells was 0% and 0%.Those results means that classical image processing is still better than neural network, this latest method could be useful in the future, but it must be improved.

Master i samfunnsernæring Background and aim: Although generally recognized that increased consumption of processed foods is contributing to the obesity epidemic, industrial processing is generally underestimated or overlooked in frameworks for studying diet and health. To close this gap in research, a new classification of food (NOVA) has recently been proposed, describing foods and dietary patterns in terms of extent and purpose of industrial processing. Research has shown that highly processed foods –“ultra-processed products”- dominate diets in high-income countries, and that these are unhealthy diets. The domination of ultra-processed products in diets has been seen in the context of the food system, characterized by a lightly regulated global food industry with profit interests in increasing sales of these products. The aim of this study is to assess food sales in Norway, applying the new classification of food. Methods: The new classification was applied to food sales data from a nationally representative sample of retailers. Foods were grouped into NOVA1 minimally processed foods, NOVA2 culinary ingredients, and NOVA3 ready-to-consume products (processed and ultra-processed products). Data from September 2005 and 2013 were analysed in Norway as a whole, in six geographical regions, and in three retail concepts. The analysis included 795 306 sales of food items. Indicators were share of purchases and expenditure for these food groups. Results: NOVA3, with more than 70% of purchases and 60% of expenditure, dominated food sales. NOVA1 accounted for 12% of purchases and 30% of expenditure. Sweets, snacks and desserts were most frequently purchased food items and accounted for the largest expenditure share both in 2005 and in 2013. Sweet ultra-processed products combined (food and beverages) accounted for every third purchase in 2013, and were purchased two and a half times more often than minimally processed food. Share of purchase and expenditure on NOVA groups changed minimally in favour of NOVA1 and in disfavour of NOVA3 between 2005 and 2013. Conclusions: The present study indicates that Norwegian diets are dominated by ready-to-consume products to an extent that is likely to be contributing to rising rates of overweight, obesity and related non-communicable diseases. Policy measures should aim at decreasing consumption of ready-to-consume products

ENGELSK: Classification of the audio track, of a motion picture, into traditional audio classes is a challenging task. The reason for this is the amount of mixed content. Speech has often background music or environmental sounds, and music has often background environmental sounds or speech. Traditional methods for separating clear audio classes have limited performance on mixed audio content. New methods and tools for automatic classification of this type of audio are therefore needed. This project investigates combinations of low level descriptors, dimensionality reduction by PCA and classification by KNN. A feature set consisting of Audio Power (AP), Audio Wave Form (AWF), Root- Mean-Square (RMS), Short Time Energy (STE), Low Short-Time Energy Ratio, Zero- Crossing Rate (ZCR), High Zero-Crossing Rate Ratio (HZCRR) in the time domain and Audio Spectrum Centroid (ASC), Fundamental Frequency (FuF), Mel, Frequency Cepstral Coefficients (MFCC), Spectrum Flux (SF) in the frequency domain is extracted on 30ms windows and integrated over a 1.2 second frame to yield a 23-dimmensional feature vector. The most suitable combination for separating speech from background music were; AP, ASC, AWF, STE, RMS, SF, fourth MFCC, AP(min scalar), ZCR(min scalar) and STE(min scalar). The combination has a majority of descriptors from the time domain. Most suitable combination of LLDs to separate speech with background environmental sounds from clear environmental sounds was found to be the 1th, 4th, 5th, 6th, 8th and 9th Mel Frequency cepstran coefficients. MFCC is in the frequency domain. Best results were achieved when the PCA returned 3 dimensions, and when the KNN classified the samples based on the 4 closest neighbors. The results from testing different mixtures of speech and music, to find the boundary where speech with background music no longer is categorized as speech, showed that the music signal had to be minimum 8 dB below the speech signal to be classified as speech. Some of the low level descriptors which traditionally performs well when separating clear classes, performed poorly in the experiments with mixed classes. Especially ZCR and FuF failed to separate speech with background music from clear music. A final experiment classifies the audio track from the motion picture ’Groundhog Day’. First 80 percent of the movie is used for training of the KNN, and the remaining 20 percent was classified. After post processing the result was 76.9 percent correctly classified. A table of content was then based on this classification.

The report was prepared for the WHO/OECD workshop on Air Quality Guidelines for Control strategies in Western and Northern Europe and include the background for selection of Air Quality Indicator (AQI). Based upon the experience gained from Norway it is concluded that one meaningful single AQI can not be developed, which combines different air pollutants and different air quality inputs. The problems of representativity might be solved by combining air quality measurements with modelling.