| In confinement swine production, environment controllers have conventionally been using air temperature as the standard, single control variable. Although the air temperature may be precisely controlled as designed, this method has its inherent drawbacks of not being interactive with the thermal needs of the pigs. Better performance of the animals may be achieved if the control is directly based on their thermal comfort behavior, which is not only affected by air temperature, but also by other thermal, social, and nutritional factors--floor type, air velocity, radiation, group size, etc.;An interactive, behavior-based control scheme was thus proposed in this dissertation. This new method explored the feasibility of using swine postural behavioral images to determine their thermal comfort state, thereby making the control decision. Namely, the animals use their own "body language" to indicate the suitability of the environment and the need for adjustment.;Experiments were conducted using early weaned pigs (13 to 16 days old) to evaluate the feasibility of automatically classifying the swine comfort behavior via image processing/analysis and neural network development. The nursery pigs were exposed to cold, comfort, and warm environments so that the corresponding postural behaviors could be induced. A total of 201 eligible behavior pictures of the pigs were scanned and digitized, and stored in computers for further analysis. Various image segmentation approaches were investigated, including histogram, thresholding, edge detection, and morphological filtering to transform the digitized gray-scale images into binary images with the pigs in white and all background in black. Conventional neural network as well as fuzzy neural net was developed to serve as the thermal behavioral classifiers, whose inputs were some features selected from the binary images. The features explored in this study consisted of Fourier coefficients, moments, perimeter and area after opening filter, and combination of the moments and the perimeter and area. Among all the features, combination of the moments and the perimeter and area gave the best classification rate for identification of the pig thermal comfort behavior. Compared with the conventional neural network, fuzzy neural networks provided membership functions-additional information that is the most useful to control strategy. Therefore, classification of swine thermal comfort behavior by image processing in conjunction with neural networks proved to be feasible. |
