Invited review: Cattle lameness detection with accelerometers.

Locomotion scoring is time consuming and is not commonly completed on farms. Farmers also underestimate their herds' lameness prevalence, a knowledge gap that impedes lameness management. Automation of lameness detection could address this knowledge gap and facilitate improved lameness management. The literature pertinent to adding lameness detection to accelerometers is reviewed in this paper. Options for lameness detection systems are examined including the choice of sensor, raw data collected, variables extracted, and statistical classification methods used. Two categories of variables derived from accelerometer-based systems are examined. These categories are behavior measures such as lying and measures of gait. For example, one measure of gait is the time a leg is swinging during a gait cycle. Some behavior-focused studies have reported accuracy levels of greater than 80%. Cow gait measures have been investigated to a lesser extent than behavior. However, classification accuracies as high as 91% using gait measures have been reported with hardware likely to be practical for commercial farms. The need for even higher accuracy and potential barriers to adoption are discussed. Significant progress is still required to realize a system with sufficient specificity and sensitivity. Lameness detection systems using 1 accelerometer per cow and a resolution lower than 100 Hz with gait measurement functions are suggested to balance cost and data requirements. However, gait measurement using accelerometers is rather underdeveloped. Therefore, a high priority should be given to the development of novel gait measures and testing their ability to differentiate lame from nonlame cows.

Temporal, spatial, inter-, and intra-cow repeatability of thermal imaging.

The objective of the present study was to quantify the within- and between-cow, operator, and day variances of various descriptive temperature parameters from different anatomical areas captured using thermal images on Holstein-Friesian cows. Three experiments were undertaken. In Exp. 1, 30 images were captured by a single operator of each of the eye, hoof, and udder from each of 45 cows; in Exp. 2, three different operators captured eye and hoof images from 12 cows; and in Exp. 3, eye and hoof images were captured by a single operator from 8 cows over a 5-d period. Maximum, minimum, and average descriptive temperature parameters were manually extracted from all thermal images within the study. The repeatability of thermal imaging and the number of replicates required to obtain a certain level of precision was evaluated. Precision was defined as the 95% CI range within which the (average of the) measured temperature(s) was expected to lie relative to the gold standard; the gold standard temperature of an entity in this study was the average of 30 temperature measurements. The partitioning of the variance into error, cow, operator, and day variances was undertaken using mixed models. Results show that the most repeatable anatomical area was the hoof, with the total proportion of variation attributed to the cow ranging from 91.37 to 99.28%. The descriptive temperature parameter with the lowest error variance was the maximum temperature for the eye (0.11°C) and udder (0.03°C) images, whereas the average temperature was the most precise descriptive temperature parameter for hoof (0.08°C) images. Additionally, no significant between-day variance was detected for maximum hoof temperatures. Results from the present study indicate that when the most precise descriptive temperature parameter is used, measurements made using infrared thermography can achieve a high level of precision in an agricultural environment if at least 3 replicate images of the eye, udder, or hooves of cows are captured and averaged. Additionally, when multiple operators capture thermal images in an agricultural environment, a standard operating procedure should be put in place to minimize the variance between operators.