This thesis is an investigation of image processing and machine learning methods to predict meat quality measures in beef cattle from ultrasound images. A method of determining two different measures of meat quality is presented, the marbling grade and the percentage of intramuscular fat. Textural features are extracted from ultrasound images of the living animals and the subsequent feature set is reduced using different methods of feature selection and the resulting feature set is used as inputs to a number of prediction algorithms. The results fro each method of feature selection and prediction methods are compared and discussed. The results of the system indicate that each of the measures can be estimated accurately; the marbling grade can be estimated to within one grade of accuracy and the percentage intramuscular fat can be estimated to within 20% of its value. Finally, the system design is discussed with suggestions of improvements to enhance the accuracy of the estimation.
A method is presented to classify the percent intramuscular fat (%IMF) for beef cattle using ultrasound imaging. As the images captured tend to include a significant amount of noise a noise reduction algorithm was used. The effectiveness of using filtered images to calculate texture measures for the classification and prediction of the %IMF is compared to the effectiveness of using unfiltered images.
A method of determining two quality measures of beef cattle using different classification networks is presented. The method involves calculating texture features from ultrasound images of the beef cattle and then predicting the final percentage intramuscular fat (IMF) and marbling grades associated with the beef cattle. This method can be used in the cattle industry to enhance current breeding techniques.
Scenes from "Where's Waldo?" are used as data sets to explore a method of pattern recognition using local histograms and neural networks. The method consists of splitting the image into a number of sub images, selecting random training images, training the neural networks and then passing the whole set of puzzles through the image to find the character Waldo. Results are shown which demonstrates that this method is effective for finding the pattern found in Waldo's shirt.