LCRM: Layer-wise Complexity Reduction Method for CNN Model Optimization on End Devices

The increasing significance of state-of-the-art convolutional neural network (CNN) models in computer vision tasks has led to their widespread use industry and academia. However, deploying these resource-limited environments, such as IoT devices or embedded GPUs, presents challenges due increased complexities resource consumption. This research paper proposes an optimization algorithm called Layer-wise Complexity Reduction Method (LCRM) address by converting accuracy-focused CNNs into lightweight models. It evaluates the standard convolution layers replaces them with most efficient combination substitutional convolutions based on output channel size. primary goal is reduce computational complexity parent hardware requirements. We assess effectiveness our framework evaluating its performance various CNN models, including AlexNet, VGG-9, U-Net, Retinex-Net, for different applications image classification, optical character recognition, segmentation, enhancement. Our experimental results show up a 95% reduction inference latency 93% energy consumption when deployed GPU. Furthermore, we compare LCRM-optimized methods, pruning, quantization, clustering, four cascaded Raspberry Pi-4. profiling experiments performed each model demonstrate that achieve comparable better accuracy than while providing added benefits 62.84% end significant memory compression reductions.