Convolutional Autoencoder Network with Multi-Instance Attention Network Based Weather Condition-Based Crop Identification in IOT

Abstract

In the field of precision agriculture, appropriate crop recommendation systems play a critical role in maximizing crop production and optimizing resource utilization. Weather condition depends on a novel approach to crop identification introduced by this research using an Internet of Things (IoT) framework with convolutional autoencoder network multi instance attention network (CAN-MIAN). This requires detailed preprocessing which also includes noise filtering, removal of missing values and duplicated data from the recommended crop dataset that includes information about rainfall, weather, agriculture, soil and fertilizers in India. Using Hybrid Crested Porcupine Zebra optimization Algorithm for feature selection will improve model accuracy. The Enhanced Sparrow Search Algorithm tuned CAN-MIAN is then used for training to enable the system predict the best crops for specific cases based on environmental parameters. As such, it provides an accurate and reliable forecast generating tools useful in agricultural decision making processes. Also, it facilitates ease in data processing flow while embedding advanced optimization algorithms into weather driven cropping pattern design through minimizing energy wastages inherent in broad acre practices. The proposed method attains higher accuracy as 97.1%.