Internet of Things (IoT) Enabled Image Segmentation Model For Lung Disease Classification: An Approach Based On Particle Swarm Optimization

In the last decades, the domain of IoT has been explored by research community due to its vast real time applications. A combination of deep learning and IoT is well accepted worldwide as using deep learning, IoT devices can be easily converted into intelligence devices. Moreover, these devices are capable enough to take the decision based on real-time data. However, deployment of deep learning model is not so easy in IoT devices as these devices are constraint with limited computational power and storage space. Generally, deep learning architectures are large in terms of storage space, and due to the complication of model, it required resources to generate the output. To overcome the storage space and the large resource barrier, we proposed the method based on the particle swarm optimization technique for compression of the UNet architecture for its easy deployment on IoT devices for semantic segmentation usages. In this paper, all the intermediate steps involved for this compression of UNet using PSO is well explained with suitable examples. Experimentally, it has been proven that the proposed algorithm compresses the UNet architecture in the chest radiograph data set by 77% after 0. 68% decrease in accuracy with an improvement in the inference time by 2.23X.