Self-Supervised Learning Graphical Neural Network Driven Prediction Model for Path-Planning and Navigation in Smart Sustainable Agriculture

摘要

Recently, autonomous systems in agriculture have garnered increased attention among research communities and industries due to their greater significance. With rapid technological advances, numerous autonomous systems have been actively deployed in agriculture to accomplish complex tasks, such as surveying agricultural fields, managing the water cycle, resolving soil compaction issues, etc. Swarm robotics, path planning, and task assignment systems play a crucial role in agricultural automation and pave the way for smart, sustainable agriculture. However, the major problem here is that this system depends on dynamic path planning activities, which are the most crucial task for autonomous robots. This is because in smart agriculture, multiple agents are associated with the system to perform various activities on their way to the destination, making the path planning actions more complicated than ever. To address these constraints, this paper presents a cost-efficient, reliable, and safe navigable swarm approach, called the Self-Supervised Learning Graphical Neural Network Driven Prediction Model (SGNN), for smart sustainable agriculture. The proposed SGNN approach enables swarm robots to select the most optimized paths, thereby reducing the distance travelled by the robots. The work’s prime focus is on developing a forecasting model for the efficient navigation and landing of swarm robots in innovative agricultural systems. The Semantic Drone Dataset is used for experimentation purposes. It is observed from the analysis that the proposed approach offers more than 85% percent accurate results in comparison to standard ResNet architectures. Additionally, this approach yields a score of around 0.8, indicating a good balance between precision and recall measures. Further, the result suggests a correlation coefficient factor of around 0.685, representing the effective classification of the target classes. Thus, the proposed SGNN approach has been experimentally proven to be more robust and efficient than existing approaches.