Detection of progression of mastitis by automatic milking systems using neuro-fuzzy networks

Abstract

Mastitis in dairy cattle is the most expensive disease in the dairy industry. It poses a significant concern to the dairy farmers as it negatively impacts both the milk yield and the cow health. Various measures have been taken to reduce the occurrence of mastitis including pre and post milking teat disinfection, good hygiene at milking time, antibiotic therapy etc. However, these measures have been found effective to a limited extent. Therefore, a need has been identified for developing a more effective, technology-based mastitis control strategy in order to have a safe, infection-free and nutritious milk supply. One potential mean of addressing the mastitis problem is through the use of intelligent computational systems such as neural and neuro-fuzzy approaches.

A number of studies using neural networks have been conducted in the past. The area of focus of these studies is the exploitation of neural network concepts such as Self Organizing Maps (SOM) that determine the health state of a cow in crisp form, i.e., whether the cow is sick or healthy. However, in many cases, a cow can be neither perfectly healthy nor clinically ill but falls in an overlapping area (e.g., marginally healthy, very ill, etc.). Also, as of now, the dynamicity of the problem hasn’t been focused on in these studies. Therefore, this study hypothesizes that a diagnosis of the health state of cattle in a more realistic fuzzy form than the current crisp form will be closer to the actual health state. This will further help in the development of effective treatment strategies leading to improved cow health, increased milk yield and substantial reduction of income loss.

This study attempts to effectively classify the health state and determine the likelihood of that health state in cows. Exploiting the data collected by robotic milking stations, the aim is to specify where the health of a cow lies in the spectrum of Mastitis. To achieve this, the existing research studies have been extended to another level by incorporating fuzziness into neural network models.

Self-organizing map is the artificial neural network being used in the study, and three different fuzzy algorithms have been applied on the results extracted from SOM. Subsequently a comparative analysis has been done to determine the fuzzy algorithm that is best suited for this particular application. The number of clusters, or health states, to be used with fuzzy algorithms has been obtained from Ward clustering of SOM. The results obtained have been justified using statistical methods. The study finally converges to a point, where a “critical area” or the “transition phase” is identified. This phase indicates the border or a “progressive” region where the healthy cattle advance towards a sickness stage. Thus, the model shows success in early identification of mastitis, which could lead to the recovery of cattle and estimation of the accurate time when the preventive measures need to be taken within the identified critical area.

The practical implementation of the research might help in saving infected dairy cattle by timely recognition of the disease and administering the treatment to infected cattle at an early stage. The quality of cattle would improve and so would the amount of milk they produce. This would further have an impact on dairy products manufactured from milk. This is not only valuable from biological perspective with enhanced longevity of cattle and maintaining diverse gene-pool but also from monetary perspective. The financial losses incurred every year, due to Mastitis disease, are of mammoth proportions. The research has the potential to contribute to a reduction of this amount significantly.