Estimating Body Condition Score in Dairy Cows From Depth Images Using Convolutional Neural Networks, Transfer Learning and Model Ensembling Techniques

Abstract

BCS (Body Condition Score) is a method to estimate body fat reserves and accumulated energy balance of cows, placing estimations (or BCS values) in a scale of 1 to 5. Periodically rating BCS of dairy cows is very important since BCS values are associated with milk production, reproduction, and health of cows. However, in practice, obtaining BCS values is a time-consuming and subjective task performed visually by expert scorers. There have been several efforts to automate BCS of dairy cows by using image analysis and machine learning techniques. In a previous work, an automatic system to estimate BCS values was proposed, which is based on Convolutional Neural Networks (CNNs). In this paper we significantly extend the techniques exploited by that system via using transfer learning and ensemble modeling techniques to further improve BCS estimation accuracy. The improved system has achieved good estimations results in comparison with the base system. Overall accuracy of BCS estimations within 0.25 units of difference from true values has increased 4% (up to 82%), while overall accuracy within 0.50 units has increased 3% (up to 97%).

An Improved Single Shot Multibox Detector Method Applied in Body Condition Score for Dairy Cows

“…Until now, there are some research achievements in assessing BCS by machine learning [14,15,16,17], they need the help of veterinarians or experts to score the cows, human empirical rules are the basis for machine learning. In reference [14] and [17], when the image data was acquiring, an expert was arranged to score the BCS of cows in situ. In our study, we made the same arrangement, two veterinary experts were asked to stand beside the milking corridor, and to record the BCS value of each cow.…”

“…A group of metrics are used to evaluate the performance of models. The following seven important terms that need to be introduced, they are also commonly used in other literature [16,17].…”

“…In 2017, Imamura et al [13] used TOF 3D cameras to evaluate BCS. Other researchers, including Rodríguez et al [14], Masahiro et al [15], Hansen et al [16], and Juan et al [17] also used 3D cameras such as Kinect to assess BCS. Machine vision methods eliminate the guesswork and the inaccuracies of manual scoring, and the efficiency has been greatly improved.…”

Deep Learning Techniques for Agronomy Applications

“…In experiments, 1661 cow images in practical experimental environments were collected and analyzed to estimate the body condition score of each cow for the evaluation of the proposed method. The results showed that both accuracy and F1-score of the proposed method were 0.97 [42].…”

“…Three papers on DL-based image recognition techniques for agronomy applications are as follows: (1) "Automatic segmentation and counting of aphid nymphs on leaves using convolutional neural networks," by Chen et al [41]; (2) "Estimating body condition score in dairy cows from depth images using convolutional neural networks, transfer learning, and model ensembling techniques," by Alvarez et al [42]; and (3) "Development of a mushroom growth measurement system applying deep learning for image recognition," by Lu et al [43].…”