An Adaptive Hybrid (1D-2D) Convolution-Based ShuffleNetV2 Mechanism for Irrigation Levels Prediction in Agricultural Fields With Smart IoTs

Rich natural resources such as fertilizers, environment, groundwater, rivers, and land are abundant in many countries. Agriculture is the primary source of income for the people living in different countries. There have not been shortages of resources like river water and groundwater, in recent decades. But, the lack of knowledge on how to use those valuable resources is the main reason for resource wastage. The amount of water applied to crop fields in a variety of soil, weather, and crop growth stages can be managed and optimized using smart farming. The crop field’s soil moisture can be measured using sensors positioned at various observation points, which will show how much water has been retained. Unfortunately, the smart farming system is not capable to receive the soil moisture data provided by the irrigation management due to issues with connectivity or sensor failure. Innovative agricultural approaches can be facilitated by the Internet of Things (IoT) technologies. These IoT nodes have encountered energy limitations and challenging routing techniques as a result of their low capacity. Therefore, it is imperative to resolve the issues by implementing an effective IoT-based irrigation system in the agricultural area. The major steps of the developed model are data collection and prediction. Initially, essential image and sensor data is attained from the benchmark resources. Next, the collected images are provided to the level of the irrigation prediction phase. This phase facilitates the farmers to maximize the crop yields and minimize the production cost. Here, effective irrigation prediction is performed using an Adaptive Hybrid (1D-2D) Convolution-based ShuffleNetV2 model (AHC-ShuffleNetV2). Moreover, the parameters of the suggested AHC-ShuffleNetV2 are optimized using a Fitness-based Piranha Foraging Optimization Algorithm (FPFOA). This increases the performance rates of the proposed model. Later, several experimental analyses are executed in the developed model over classical techniques to display their effectualness rate. When considering the sigmoid activation function, the implemented smart irrigation level prediction framework’s RMSE was minimized by 73.15% of POA-ShuffleNetV2, 72.36% of RSA-ShuffleNetV2, 78.94% of MRS-ShuffleNetV2, and 79.47% of PFOA-ShuffleNetV2 respectively. Hence, it is revealed that the designed smart irrigation level prediction model attained low error rates and also achieved higher efficacy than the other baseline techniques.