Automatic Lameness Detection in a Milking Robot : Instrumentation, measurement software, algorithms for data analysis and a neural network model

The aim of this thesis is to develop a fully automatic lameness detection system that operates in a milking robot. The instrumentation, measurement software, algorithms for data analysis and a neural network model for lameness detection were developed. Automatic milking has become a common practice in dairy husbandry, and in the year 2006 about 4000 farms worldwide used over 6000 milking robots. There is a worldwide movement with the objective of fully automating every process from feeding to milking. Increase in automation is a consequence of increasing farm sizes, the demand for more efficient production and the growth of labour costs. As the level of automation increases, the time that the cattle keeper uses for monitoring animals often decreases. This has created a need for systems for automatically monitoring the health of farm animals. The popularity of milking robots also offers a new and unique possibility to monitor animals in a single confined space up to four times daily. Lameness is a crucial welfare issue in the modern dairy industry. Limb disorders cause serious welfare, health and economic problems especially in loose housing of cattle. Lameness causes losses in milk production and leads to early culling of animals. These costs could be reduced with early identification and treatment. At present, only a few methods for automatically detecting lameness have been developed, and the most common methods used for lameness detection and assessment are various visual locomotion scoring systems. The problem with locomotion scoring is that it needs experience to be conducted properly, it is labour intensive as an on-farm method and the results are subjective. A four balance system for measuring the leg load distribution of dairy cows during milking in order to detect lameness was developed and set up in the University of Helsinki Research farm Suitia. The leg weights of 73 cows were successfully recorded during almost 10,000 robotic milkings over a period of 5 months. The cows were locomotion scored weekly, and the lame cows were inspected clinically for hoof lesions. Unsuccessful measurements, caused by cows standing outside the balances, were removed from the data with a special algorithm, and the mean leg loads and the number of kicks during milking was calculated. In order to develop an expert system to automatically detect lameness cases, a model was needed. A probabilistic neural network (PNN) classifier model was chosen for the task. The data was divided in two parts and 5,074 measurements from 37 cows were used to train the model. The operation of the model was evaluated for its ability to detect lameness in the validating dataset, which had 4,868 measurements from 36 cows. The model was able to classify 96% of the measurements correctly as sound or lame cows, and 100% of the lameness cases in the validation data were identified. The number of measurements causing false alarms was 1.1%. The developed model has the potential to be used for on-farm decision support and can be used in a real-time lameness monitoring system.

Statistical and Information-Theoretic Methods for Data-Analysis

In this Thesis, we develop theory and methods for computational data analysis. The problems in data analysis are approached from three perspectives: statistical learning theory, the Bayesian framework, and the information-theoretic minimum description length (MDL) principle. Contributions in statistical learning theory address the possibility of generalization to unseen cases, and regression analysis with partially observed data with an application to mobile device positioning. In the second part of the Thesis, we discuss so called Bayesian network classifiers, and show that they are closely related to logistic regression models. In the final part, we apply the MDL principle to tracing the history of old manuscripts, and to noise reduction in digital signals.

Geant4 Hadronic Cascade Models and CMS Data Analysis : Computational Challenges in the LHC era

This work belongs to the field of computational high-energy physics (HEP). The key methods used in this thesis work to meet the challenges raised by the Large Hadron Collider (LHC) era experiments are object-orientation with software engineering, Monte Carlo simulation, the computer technology of clusters, and artificial neural networks. The first aspect discussed is the development of hadronic cascade models, used for the accurate simulation of medium-energy hadron-nucleus reactions, up to 10 GeV. These models are typically needed in hadronic calorimeter studies and in the estimation of radiation backgrounds. Various applications outside HEP include the medical field (such as hadron treatment simulations), space science (satellite shielding), and nuclear physics (spallation studies). Validation results are presented for several significant improvements released in Geant4 simulation tool, and the significance of the new models for computing in the Large Hadron Collider era is estimated. In particular, we estimate the ability of the Bertini cascade to simulate Compact Muon Solenoid (CMS) hadron calorimeter HCAL. LHC test beam activity has a tightly coupled cycle of simulation-to-data analysis. Typically, a Geant4 computer experiment is used to understand test beam measurements. Thus an another aspect of this thesis is a description of studies related to developing new CMS H2 test beam data analysis tools and performing data analysis on the basis of CMS Monte Carlo events. These events have been simulated in detail using Geant4 physics models, full CMS detector description, and event reconstruction. Using the ROOT data analysis framework we have developed an offline ANN-based approach to tag b-jets associated with heavy neutral Higgs particles, and we show that this kind of NN methodology can be successfully used to separate the Higgs signal from the background in the CMS experiment.

Accelerator Mass Spectrometry and Bayesian Data Analysis

Accelerator mass spectrometry (AMS) is an ultrasensitive technique for measuring the concentration of a single isotope. The electric and magnetic fields of an electrostatic accelerator system are used to filter out other isotopes from the ion beam. The high velocity means that molecules can be destroyed and removed from the measurement background. As a result, concentrations down to one atom in 10^16 atoms are measurable. This thesis describes the construction of the new AMS system in the Accelerator Laboratory of the University of Helsinki. The system is described in detail along with the relevant ion optics. System performance and some of the 14C measurements done with the system are described. In a second part of the thesis, a novel statistical model for the analysis of AMS data is presented. Bayesian methods are used in order to make the best use of the available information. In the new model, instrumental drift is modelled with a continuous first-order autoregressive process. This enables rigorous normalization to standards measured at different times. The Poisson statistical nature of a 14C measurement is also taken into account properly, so that uncertainty estimates are much more stable. It is shown that, overall, the new model improves both the accuracy and the precision of AMS measurements. In particular, the results can be improved for samples with very low 14C concentrations or measured only a few times.

Residual noise covariance for Planck low-resolution data analysis

Aims: Develop and validate tools to estimate residual noise covariance in Planck frequency maps. Quantify signal error effects and compare different techniques to produce low-resolution maps. Methods: We derive analytical estimates of covariance of the residual noise contained in low-resolution maps produced using a number of map-making approaches. We test these analytical predictions using Monte Carlo simulations and their impact on angular power spectrum estimation. We use simulations to quantify the level of signal errors incurred in different resolution downgrading schemes considered in this work. Results: We find an excellent agreement between the optimal residual noise covariance matrices and Monte Carlo noise maps. For destriping map-makers, the extent of agreement is dictated by the knee frequency of the correlated noise component and the chosen baseline offset length. The significance of signal striping is shown to be insignificant when properly dealt with. In map resolution downgrading, we find that a carefully selected window function is required to reduce aliasing to the sub-percent level at multipoles, ell > 2Nside, where Nside is the HEALPix resolution parameter. We show that sufficient characterization of the residual noise is unavoidable if one is to draw reliable contraints on large scale anisotropy. Conclusions: We have described how to compute the low-resolution maps, with a controlled sky signal level, and a reliable estimate of covariance of the residual noise. We have also presented a method to smooth the residual noise covariance matrices to describe the noise correlations in smoothed, bandwidth limited maps.

Identifying Meaningful Places

Place identification refers to the process of analyzing sensor data in order to detect places, i.e., spatial areas that are linked with activities and associated with meanings. Place information can be used, e.g., to provide awareness cues in applications that support social interactions, to provide personalized and location-sensitive information to the user, and to support mobile user studies by providing cues about the situations the study participant has encountered. Regularities in human movement patterns make it possible to detect personally meaningful places by analyzing location traces of a user. This thesis focuses on providing system level support for place identification, as well as on algorithmic issues related to the place identification process. The move from location to place requires interactions between location sensing technologies (e.g., GPS or GSM positioning), algorithms that identify places from location data and applications and services that utilize place information. These interactions can be facilitated using a mobile platform, i.e., an application or framework that runs on a mobile phone. For the purposes of this thesis, mobile platforms automate data capture and processing and provide means for disseminating data to applications and other system components. The first contribution of the thesis is BeTelGeuse, a freely available, open source mobile platform that supports multiple runtime environments. The actual place identification process can be understood as a data analysis task where the goal is to analyze (location) measurements and to identify areas that are meaningful to the user. The second contribution of the thesis is the Dirichlet Process Clustering (DPCluster) algorithm, a novel place identification algorithm. The performance of the DPCluster algorithm is evaluated using twelve different datasets that have been collected by different users, at different locations and over different periods of time. As part of the evaluation we compare the DPCluster algorithm against other state-of-the-art place identification algorithms. The results indicate that the DPCluster algorithm provides improved generalization performance against spatial and temporal variations in location measurements.

Data fusion and matching by maximizing statistical dependencies

The core aim of machine learning is to make a computer program learn from the experience. Learning from data is usually defined as a task of learning regularities or patterns in data in order to extract useful information, or to learn the underlying concept. An important sub-field of machine learning is called multi-view learning where the task is to learn from multiple data sets or views describing the same underlying concept. A typical example of such scenario would be to study a biological concept using several biological measurements like gene expression, protein expression and metabolic profiles, or to classify web pages based on their content and the contents of their hyperlinks. In this thesis, novel problem formulations and methods for multi-view learning are presented. The contributions include a linear data fusion approach during exploratory data analysis, a new measure to evaluate different kinds of representations for textual data, and an extension of multi-view learning for novel scenarios where the correspondence of samples in the different views or data sets is not known in advance. In order to infer the one-to-one correspondence of samples between two views, a novel concept of multi-view matching is proposed. The matching algorithm is completely data-driven and is demonstrated in several applications such as matching of metabolites between humans and mice, and matching of sentences between documents in two languages.