Optimized MobileNet Architecture for Accurate and Efficient Wildlife Classification in IoT Applications

The growing demand for real-time, resourceefficient deep learning models has motivated the exploration of lightweight architectures for various applications. This study proposes MobileNet as an alternative to InceptionResNetV2 for wildlife image classification in IoT-enabled systems. MobileNet efficiency stems from its use of depthwise separable convolutions, which reduce computational overhead and make it ideal for devices with limited resources. A comparative analysis of the two algorithms using a dataset of wildlife images. Results indicate that MobileNet achieves a 6.13% higher training accuracy and an 81.11% reduction in training loss compared to InceptionResNetV2. Both models attained perfect validation accuracy, but MobileNet exhibited a slightly higher validation loss, which remained negligible. These findings demonstrate MobileNet ability to balance efficiency and accuracy, making it a superior choice for deployment in edge devices. This work highlights the potential of lightweight architectures in IoT ecosystems, providing a scalable and costeffective solution for real-time image classification.