Real time expert system for detecting object region and working state of aerators
share
Aerators are essential and important auxiliary devices in intensive culture, especially in industrial culture of aquaculture. Using existing surveillance cameras to realize real-time and automatic detection of the object region and working state of aerators is a no-cost and real-time method, which also provides a useful reference and attempt for intelligent agriculture and agricultural expert systems. However, this expert system mainly faces three challenges: Firstly, the size and shape of object region vary greatly depending on the position and angle of surveillance cameras. Secondly, the interference factors, e.g., illumination, occlusion, complex background, etc., are strong, so feature extraction is difficult to guarantee robustness. Finally, the expert system in real applications requires real-time, robust and accurate features. To tackle these aforementioned challenges, we propose an object region detection process based on reference frame Kanade Lucas Tomasi (RF-KLT) algorithm to screen the small object region from candidate regions. Moreover, we present a robust motion feature extraction method based on RF-KLT algorithm in a fixed region. In addition, we also introduce a dimension reduction method for time series, which is used to establish a feature dataset with obvious boundaries between classes. The experimental results show that the accuracy for detecting object region and working state of aerators in the complex background is 100 second (FPS) according to the different types of surveillance camera. The proposed expert system presents a complete solution for the object region and working state detection of aerators, and achieves real-time, online and state-of-the-art results in both the actual dataset and the augmented dataset.
READ FULL TEXT