Precision Agriculture Science and Technology ISSN:2672-0086(Print) 2713-5632(Online) 정밀농업과학기술

In this study, we purposed autonomous combine harvester’s environment detection system based on deep neural networks. We conducted to develop detection system in order to comfortably apply to combine harvester without structural and design modifications. Images was acquired during combine harvesting operation, model architecture was constructed 4-layers convolutional neural networks for rough and fast execution. Area detection was performed for harvestable areas and obstacle areas, detected area’s decision boundaries was established after thresholding process to facilitate information transmission to the higher-level controller. The results of classification accuracy were observed 99.3% in harvestable area, and 91.9% in obstacle area. The environment detection speed was measured between 25hz and 40hz, thus fps (frame per second) was observed at a 30fps. From this result, environment detection could be adopted to autonomous combine harvester without structural and design modifications, and through 4-layer convolutional neural networks model, target area for autonomous combine’s environment detection was easily identified. However, this research is on rapid delivering of detection information in autonomous combine harvester, and it is considered that it can be extended to various application domains through hardware optimization and post-processing algorithms.