Detection of cattle diseases with image processing methods

Today, the need for image processing methods in veterinary research is increasing day by day. This need becomes even more important with the need for image processing methods, especially in the early diagnosis of cattle diseases. It may take time for the owner of the animal to consult a veterinarian, especially in terms of transportation, in case the cattle get sick, and the disease may progress until the diagnosis is made by the veterinarian. For this reason, it is necessary to develop software that will provide suggestions for the early diagnosis of diseases of cattle species, especially with image processing methods. In this thesis, the detection of various cattle diseases with the help of image classification, which is one of the artificial intelligence methods, was carried out. In this study, before creating the artificial intelligence model, images were collected, diseases were analyzed, classified on the data and a data set was created. In order to increase the model performance and eliminate the imbalance in the dataset, preprocessing processes were performed on the images in the dataset and the marked image data with histogram equalization, data augmentation and SMOTE methods. After the dataset was prepared, classification model training was carried out and performance analyzes were made. As a result of the performance analysis, it was decided that the most suitable model for the problem was the model trained with MobileNet-V3, which gave the highest performance with 99.94% accuracy, 99.94% sensitivity, 99.94% sensitivity and 99.99% AUC score and 0.007 loss, and the resulting weight file was sent to the prepared web service. has been integrated. Later, a mobile application was prepared with Flutter, and disease detection was carried out via the web service and transmitted to the mobile application. The mobile application prepared in this study offers veterinarians and cattle breeders a new perspective and convenience in disease diagnosis by showing the effective use with high classification performance obtained after training current artificial intelligence methods within the scope of preliminary diagnosis of cattle diseases.