Training deep neural networks for wireless sensor networks using loosely and weakly labeled images

Although deep learning has achieved remarkable successes over the past years, few reports have been published about applying neural networks to Wireless Sensor Networks (WSNs) for image targets recognition where data, energy, computation resources are limited. In this work, a Cost-Effective Domain Generalization (CEDG) algorithm proposed train an efficient network with minimum labor requirements. CEDG transfers from publicly available source domain application-specific target through automatically allocated synthetic domain. The is isolated parameters tuning and used model selection testing only. significantly different because it new categories consisted of low-quality images that out focus, low in resolution, illumination, photographing angle. trained 7M (ResNet-20 41M) multiplications per prediction small enough allow digital signal processor chip do real-time recognitions our WSN. category-level averaged error on unseen unbalanced decreased by 41.12%.