Real-time early corn and weed classification using hybrid visual features based on parallel processing and tracking

Corn and weed classification using video data is required to run in real-time. Video data processing is necessary to consider its speed and correct classification. Problems arise when the classification speed does not meet the video's frame per second (fps) specifications. This research proposes a combination of parallel processing and tracking to increase the speed of video data processing. The speed of video processing is also supported by the machine learning method. The combination of visual features through feature selection is also applied in this research. The results of the non-parallel experiment produced an average of 25.45 fps, while the proposed method produced 43.99 fps or an increase of 72.82%. This speed increase is due to parallel processing through MPI4Py at the pre-processing and color segmentation stages, which optimizes multi-core CPU resources. In addition, the feature extraction and classification stages also use skip frames so that not all frames are processed. The corn and weed classification results experienced a slight decrease. This decrease in classification results is not as significant as the resulting processing speed. Through these results, the proposed method enables real-time processing, making it useful for video data processing.