Abstract:
Objective To quickly, accurately and efficiently obtain the information of the number of cotton plants in the seedling stage using consumer-grade UAV, and expand the application scenarios of low-cost UAV in smart agriculture.
Methods An interaction experiment of 5 planting densities and 3 varieties was designed. The visible light images of UAV at the 3−4 leaf stage of each treatment were acquired by the AIR 2S UAV at a flight altitude of 15 m. Eight color indices and two Otsu's methods were calculated to extract the cotton seedling targets; Morphological operations and positional relationships were used to optimize the binary images and remove noise; 19 morphological parameter features of cotton seedlings were extracted, key features were screened, and the estimation models of the number of cotton seedlings were constructed using decision tree, random forest and support vector machine algorithms.
Results The random forest estimation model of the number of cotton seedlings constructed based on 9 key features had the highest fitting degree between the estimated number of plants and the actual number of plants, with an R2 of 0.997, an RMSE of 1.049, and the accuracy rate, recall rate and precision rate being 99.12%, 99.37% and 99.75% respectively.
Conclusion The combination of consumer-grade UAV and open-source image analysis algorithms can achieve rapid, accurate and economical monitoring of cotton seedling conditions. This low-cost and easy-to-operate means of monitoring seedling conditions can be extended to more crops and scenarios, bringing a more convenient smart agriculture management experience to the vast number of front-line users.
下载: