Contents I. Introduction
Every species has its different kinds of breeds. For instance, domestic dogs have golden retrievers, labradors, huskies, poodles, and more. Hamsters as species, also have different kinds of breeds, but based on their physical appearance, they may be difficult to distinguish at first glance especially for the people who know less about hamsters. Unlike dogs, their breeds are easily discernible because of certain factors such as fur color, facial structure, body shape, size, and their behavior. An existing research, implemented a system designed to recognize dogs through their nose prints as biometrics [1]. In the case of hamsters, usually, one can tell the breed by looking at the patterns in the hamster's facial structure aside from observing their size and fur color. So, one must be knowledgeable first for people to tell the breed of a hamster. In the subject of keeping them as pets, the domestication of hamsters became a trend due to how easy they were to keep and breed. In contrast to their sightings in pet shops and labs, their sightings in the wild have been steadily declining due to industrialization and urbanization. One of the most common pet breeds, Syrian hamsters are now considered vulnerable in the wild nearing endangerment due to the risks of these environmental factors [2]. Hamsters are not that adaptable to their environment in comparison to other domesticated animals. These domesticated hamsters are kept in captivity and are usually bred in herds in pet shops and livestock facilities for preservation. Their environment in these facilities is less than ideal as the effect of these solitary creatures being confined in a single cage together places stress on these animals. Living conditions of these domesticated animals can improve by educating pet owners about their different dietary needs and to enrich the environment of a specific breed of hamster suitable to their natural behavior [3]. The convolutional neural network is a kind of multi-layer perceptron which is designed for two-dimensional image recognition as a deep learning algorithm, and each kernel in different layers is learned spontaneously [4]. CNN offers a higher success rate for image recognition than other image classification methods due to its proficiency in image representation [5], which explains a lot of research involving image processing and recognition utilizing CNN. A related study concluded that CNN was able to reach a high accuracy rating for determining sample damage types in corn [6]. CNN was also able to classify and count White Blood Cells based on microscopic blood images [7]. Another study utilized CNN for detection of corn leaf diseases with the implementation of OpenMP which was a main factor for attaining high accuracy in classifying leaf diseases with Raspberry Pi [8]. The performance of the whole network can be improved by increasing the number of samples. CNN is mostly used in studies that involve image recognition [9]–[15]. While gathering and studying related literatures, the researchers have found out that there is no image capturing system designed for identifying different breeds of hamsters yet. Although previous studies such as the rare animal image recognition based on CNN used CNN to train the system to identify rare animals only [4]. Specifically, a VGG16 architecture is used for the system. Among the various deep learning architectures available for image classification, VGG16 has become one of the most popular choices [16]. Studies that are related to image detection, recognition, or classification have used VGG16 as their main architecture [17]–[20] The main objective of this research is to identify different hamster breeds using transfer learning method and Convolutional Neural Networks. The specific objectives are as follows: (1) to develop an image capturing system that will capture images of different hamsters, (2) to utilize Convolutional Neural Network for the device to identify specific patterns that would lead to the recognition of the details on hamster breeds via Raspberry Pi 4, and (3) to evaluate the findings of the system using Confusion Matrix. The research will provide insights in recognizing hamster breeds for pet owners and breeders to understand a hamster's traits and quirks by their genealogy. Through this research will be able to extrapolate information on a particular breed of hamster's behavior. The research may also help pet owners to prevent inbreeding as well as prevent health risks that are prone to specific breeds. The dataset that will be used to train the program will only cover images of hamsters that are kept as pets, it will not cover and recognize species that are found in the wild or animals in the same order of rodents. Four breeds will only be included in the dataset: (1) Campbell, (2) Roborovski, (3) Syrian, (4) Winter White. And in each breed, the researchers will only collect 600 images, making it a total of 2400 images. The images will only come from the researchers themselves and some of the images will be taken from the web. Lastly, the main hardware that will be used is Raspberry Pi 4 Model B with an 4GB memory.
