The use of mechanical brushes is a strategy for improving the welfare of dairy cows; however, measuring brush usage is a complex and time-consuming task. This study aimed to develop an artificial neural network (ANN) to identify dairy cows’ usage of mechanical brushes using image capture. The database (503 images) was built with publicly available images (e.g., Google Images and YouTube) and manually divided into two folders: 1 - "Using Brush" (251 images) and 2 - "Not Using" (252 images). The ANN performed image resizing, image preprocessing, and convolutional training over 10 epochs with 16 steps for data processing and machine learning, resulting in a total data loss of 0.0074 with an accuracy of 0.998. ANN algorithms and image processing software were implemented using the Python programming language and the open-source libraries TensorFlow, Keras, and Python Screen Capture. A data augmentation strategy (e.g., rotation and filters) was used to build the test datasets. ANN validation was performed based on its errors and successes in three distinct tests, resulting in a precision of 0.75, recall 0.62, and an average processing time of 40 milliseconds per image. In a farm environment, where cows need to be continuously monitored, a moderate precision of 0.75 can allow for the detection of relevant behaviors, such as the frequency of brush use. This information can be used to evaluate and improve animal welfare by ensuring cows are using the brushes as intended, which can lead to reduced stress and increased comfort. Monitoring brush usage may also offer insights into cow health and productivity. Our findings indicate that the developed convolutional learning neural network has the capacity to contribute to the stufy of animal behavior and assist in identifying the usage of mechanical brushes by dairy cows, as it demonstrated high processing speed and accuracy.