Estimation of rapeseed crop density at the two-leaf stage based on image processing and using a combination of Gaussian Mixture Model and k-means clustering algorithms

Today, with the expansion of industrial agriculture, the use of new sciences and technologies such as artificial intelligence has a significant role in increasing productivity and smartening of agricultural methods. These methods include estimating seedling density using image processing methods. In addition to helping agricultural land management, this is also effective in assessing the amount of fertilizer and chemicals and determining the storage space required. In this paper, a new method for estimating rapeseed crop density at the two-leaf stage is proposed. To prepare the images, first, images were taken from several random areas of the rapeseed field in the two-leaf stage located in the village of Mohammadabad, Qazvin, using square frames one meter long to meet the usual standards; Then the Gaussian mixed model with the Expectation Maximization algorithm is used to segment the images. In order to improve the segmentation of the images, the k-mean clustering algorithm was used and finally, when the leaves were separated from the other components of the image, morphological operators were used to count the number of products in the images. The number of rapeseed products in the images is also averaged manually and used as reference values to evaluate the performance of the proposed algorithm. The results of the proposed method have a correlation of R=0.96 with the manual counting method and have an accuracy of 96.5%. The results of the proposed method are also compared with two common methods called the Normalized Difference Index (NDI) and the Otsu threshold methods which are based on the color characteristics of the images and used in recent studies, and it is observed that the proposed method works better. Although the images were taken in different environmental conditions and with different light intensities, the error rate of the proposed method for the images used was only less than four percent, which shows the efficiency of the proposed method in estimating canola density. Therefore, the proposed method can be used in estimating canola seedlings in practice in agricultural fields.