Identification of Chicken Eimeria Species from Microscopic Images by Using MLP Deep Learning Algorithm

Abstract

Eimeria has more than one species of every single genus of animals that causes diseases that may spread at fast speed and therefore adversely affects animal productivities and results in animal death. It is therefore essential to detect the disease and prevent its spread at the earliest stage. There have been some attempts to address this problem through the analysis of microscopic images. However, due to the complexity, diversity, and similarity of the types of the species, there need more sophisticated methods to be adapted for the intelligent and automated analysis of their microscopic images by using machine- learning methods. To tackle this problem, a deep-learning-based architecture has been proposed and successfully adapted in this study where Chicken fecal microscopy images have been analyzed to identify nine types of these species. The methodology developed includes two main parts, namely (i) pre-processing steps include the techniques that convert image into gray level, extract cell walls, remove background, rotate cell to vertically aligned position and move to their center and (ii) MLP-based deep learning technique to learn features and classify the images, for which Keras model was utilized. Based on the outcome of a 5-fold cross validation that was repeated for 100 times, the approach taken has yielded an average accuracy of 83.75%±0.60, which is comparable to the existing methods.