| Popis: |
Computer vision is an advanced technology and is widely used in aspects such as IoT, agricultural robotics and machine learning. The detection and extraction of plants are the primary tasks of this application. This article shows several index-based colour analysis computer vision methods and relative improvement to extract the plants from backgrounds which contain complex information. The sample images are from the Plant Phenotyping Dataset which contains arabidopsis as desired plants and interference information (metal, moss, white table, etc.). A comparison of the three methods is provided. These methods are Excess Green Index (ExG), Excess Red Index (ExR), and the combination of ExG and ExR (ExG-ExR). Meanwhile, a mask to remove the non-green object (BR mask), Gaussian filter, and erosion operation are added as an improvement to remove the remains. With the evaluation method such as Intersection over Union (IoU) and S⊘rensen-Dice Similarity Coefficient (DSC), the combination of ExG-ExR, BR mask, Gaussian filter, and erosion shows the best performance with IoU and DSC rate 0.8818 and 0.9368 respectively. However, one setback is that green information like moss is hard to be distinguished from desired plants with these methods. This problem will be researched in the future. |
