Fast discrimination of female and male pigeon eggs using internet of things in combined with Vis-NIR spectroscopy and chemometrics

[Display omitted] •IoT framework is established for discrimination of pigeon eggs by sensor detection.•NIR spectroscopy and its chemometric methods are studied for cloud computing.•A modified RWNN architecture is designed for intelligent model optimization.•Adaptive learning strategy is proposed to refine the network and its hyperparameters. In livestock industry, the female and male pigeons have different follow-up functions. The discrimination of female and male pigeons is an intensive concern for breeding tasks. In daily cultivation, the livestock staffs cannot distinguish the pigeon sex until the child pigeon is born. This lag in judgment seriously affects the freshness of pigeon eggs and timely sales plans. To solve this problem, we construct an internet of things (IoT) framework for modeling to discriminate a batch of pigeon eggs based on the instant data detection by visible and near-infrared (Vis-NIR) spectroscopy technology. In practice, the spectral detection data is monitored by multi-locational Vis-NIR sensors and immediately delivered to the cloud unit of the IoT framework. A random weight neural network (RWNN) architecture is designed as the intelligent computing module for model training and optimization, so that the cloud unit is able to deal with the constant inflow of Vis-NIR big data. An adaptive learning strategy is also designed to tune the network linkage weights as well as relevant hyperparameters. Partial least squares discriminant analysis is embedded in the Softmax unit for model discrimination, to optimize data processing with spectral properties. Experimental results proves that the adaptive RWNN architecture is able to observe high prediction accuracy when modeling on the early 5th-, 6th-, 7th- and 8th- hatching days for the distinguishment of the female and male pigeon eggs. Thus, the IoT-based Vis-NIR technology is prospectively expected to process the online big data in support with the adaptive RWNN modeling architecture.