Real-Time Cat Detection System using MobileNet-SSD V2

The development of a cat-repellent system needs a deep learning-based object detection method to detect cats, as the current system only uses PIR sensors and cannot identify specific kinds of objects. The system detects the presence of cats using the MobileNet-SSD v2 model, a combination of MobileNet v2 and Single Shot Detector (SSD). The model’s performance is tested using the parameters of precision, recall, and loss. The Node-Red platform and the TensorFlow.js library are used to implement models on Raspberry Pi 4. Based on the results and analysis, the MobileNet-SSD v2 model creation was successful, with an accuracy value of 0.91 and a recall value of 0.89. The resulting F1-score metric of 0.92 shows a balance between precision and recall. The test results indicate that the accuracy of the model created by the transfer learning method using the pre-trained MobileNet-SSD v2 model is 82%. However, environmental factors such as distance, brightness, and camera resolution can also impact the model’s efficacy. This model can be integrated with the ultrasonic and water sprayer sub-system to enhance cat repellents’ efficacy. Future research can compare the object detection model with other models to increase detection speed and accuracy.