Efficient herd – outlier detection in livestock monitoring system based on density – Based spatial clustering

Abstract

In today's society, increasing the quality and the productivity of dairy products are very important and need detailed data collection and analysis. Manual collection of data and its analysis for livestock monitoring is costly in terms of high man power and time consumption. In order to overcome this deficit, object detection and clustering methods are investigated in this research as it is in line with Smart Farming 4.0. Faster RCNN is used to help ranchers to detect livestock while clustering methods help to detect the herds and outliers effectively and efficiently. In clustering methods, K-means clustering technique and Density-Based Spatial Clustering of Application with Noise or DBScan clustering technique are adopted. In K-means clustering, k is an important parameter which represents the number of clusters. By changing the number of clusters, the pattern of clusters is observed. Then, the best k value is selected. In DBScan clustering, epsilon is an important parameter which represents the circle radius from a particular data point. The higher the value of epsilon, the formation of clusters becomes easier as it is easy to accept data point in a larger circle radius to form cluster. By changing the epsilon, the pattern of cluster is observed and chosen. Euclidean distance and Manhattan distance are used to compare the effects of different distance metrics on the results of clusters. Cluster pattern is compared between K-means and DBScan techniques. Obtained results show that DBScan overwhelmed K-means in term of efficient clustering in detecting the herds and outliers of livestock.