Detection of watermelon in RGB images via unmanned aerial vehicle by utilising texture features for predicting yield

Abstract

Unmanned aerial vehicles equipped with a digital camera are one of technologies that the precision agriculture takes advantage. In this study conducted in 2020, watermelons in the images obtained by an unmanned aerial vehicle from the watermelon field, which is in Sarıçam, Adana, Turkey, were segmented. The original image was processed in two ways. First, the image was converted to grayscale and then divided into blocks by an overlapping sliding window. The grey level co-occurrence matrixes from these blocks were obtained and six Haralick texture features were computed. Then, blocks were classified to one of three categories, soil, leaf and watermelon, by employing a multiclass support vector machine with radial basis kernel. Meanwhile, the original image was partitioned into three groups by using k-means clustering and the group having the highest blue component at its centre was selected. Finally, the two outcomes were fused to obtain possible watermelon regions in the image. The average categorization accuracy and the rate of detected watermelons without incorporating clustering outcome were 96.51% and 98.50% respectively. The watermelon detection performance was enhanced by fusing the clustering result