Domesticated Cat Activity Detection Using Deep Learning Technique

Cats are one of the most well-known pet species worldwide. Cat owners take great care to ensure their feline companions' safety. Continuous observation of cats' behavior can help to assure their welfare. Due to the substantial benefits of sensor technology in the recent era, studies about activity detection have become very prominent. Applications for automated cat monitoring include monitoring and surveillance systems that accurately identify pets using deep learning methods for classifying pet activities. We can now use cutting-edge ways to get better results thanks to the technological revolution. Recently, Long Short-Term Memory (LSTM) has worked as a cutting-edge methodology for anomaly detection in manufacturing environments, sensor-based health monitoring systems, and signal processing-based systems. As compared to other models employed in the study for categorizing various types of activities, the CNN-LSTM architecture is seen to be the most effective. Additionally, it was discovered that all models outperformed their counterparts in data from the accelerometer, gyroscope, and magnetometer. The goal of this study is to use multi-axis sensor devices to identify ten pet behaviors, such as walking, running, eating, jumping, resting, etc. The approach for gathering data for this study involved 10 cats of various breeds, sexes (male=6, female=4), ages (age = 43), and the cats are small, medium, and large-sized in a safe setting. Following the data collection, the stages of data synchronization and data preparation were taken into consideration to eliminate unnecessary data from the dataset. We employed the class-weight technique to solve the dataset's imbalance issues. We used the class weight method to apply the LSTM algorithm. 99.45% training and 96.23% validation accuracy were displayed by the model using the class weight technique. The LSTM method will be useful for tracking cat behavior and for real-time activity monitoring.