Table 1.
Inputs and outputs of computer vision algorithms used in cattle with their respective functions
| Input | Output | Applicable function in cattle | Research example |
|---|---|---|---|
| Image | Classification probabilities indicating different classes or situations | Classification of animal behavior (eating, resting, biting, chewing) and classification of health condition (e.g., lameness) | Rodríguez Alvarez et al. (2019) |
| Image | Regression of numerical values as a result of predictions (supervised learning) | Estimation of numerical values such as BW, dry matter intake, height, milk yield, or average daily gain | Gjergji et al. (2020) |
| Image | Segmentation probabilities for each pixel according to its corresponding class | Classification objects/items within images, such as collars, ear tags, health damages | Wu et al. (2020) |
| Image | Combinations of regression (e.g., bounding box coordinates) and probabilities (e.g., Mask RCNN with instance pixel masks within the bounding boxes) | A lightweight and high-precision detection model based on the YOLOv4 framework, named GG-YOLOv4, is used to automatically detect ocular surface temperatures from the thermal images of dairy cows | Wang et al. (2022) |
| Image or Video | 3D reconstruction with point clouds | An unsupervised DBSCAN clustering algorithm was proposed to calibrate the leg region boundary based on clustering features | Li et al. (2022) |
| Video | Classification probability for sequential frames (e.g., drinking, estrus) | An algorithm for tracking the beef cattle’s key body parts, such as head–ear–neck position, using a state-of-the-art deep learning architecture, DeepLabCut. The extracted key points were analyzed using an extended short-term memory model to classify drinking and non-drinking periods | Islam et al. (2023) |
| Video | Combination of classification, regression, object tracking ID, and downstream analysis | A CNN model, which included a tensor of 4-channel matrices of data, each with 480 × 640 pixels. The model design was inspired by ResNet CNN, which achieved the best results in an ILSVRC classification and detection competition | Bezen et al. (2020) |
Abbreviations: CNN, convolutional neural networks; RCNN, region-based convolutional neural networks; YOLO, you only look once.