Abstract: Abstract: Accurate identification of maize stubble row has widely dominated the automatic row-followed seeding using machine vision. However, it is difficult to segment the images of stubble row in the maize stubble field harvested by combine harvesters, due mainly to the indistinct chromaticity difference with naked land surface and maize residues. In this study, image segmentation was presented using a support vector machine (SVM), in order to realize precise and rapid segmentation of the maize stubble row. Firstly, principal component analysis (PCA) was used for dimensionality reduction and feature optimization of the dataset, where the specific features were selected to distinguish standing stubble, naked land surface, and maize residues. Especially, the 1 500 sample images of standing stubble, 1500 sample images of the naked land surface, and 1 500 sample images of maize residues were collected, while, 2 210 features containing 697 color features, and 1 513 texture features were obtained using sample images. Then, PCA was used to choose 21 color features of the standing stubble, naked land surface, and maize residues in the R, G, B, L, a, b, v, YIQ-V and YCbCr-Y components from the datasets. The selected color features were constructed into a 21-dimensional feature vector, which was used as the input of the standing stubble SVM recognition model. Secondly, the region of interest (ROI) was selected in the middle of the image with the integrated maize stubble row for the higher efficiency of image segmentation. Finally, the trained SVM recognition model was used for the slide detection of standing stubble within the ROI with a window of 25×25(pixel). If the currently detected window was standing stubble in slide detection, the grayscale value would be set to 255. The maize stubble row was segmented by a threshold when the slide detection was complete. The segmented binary image was then optimized using the morphological open operation processing with a disc-shaped structural element with a radius of 2 pixels. Furthermore, 100 test images were collected to verify the segmentation performance from the Scientific Observing and Experimental Station of Arable Land Conservation (North Hebei), Ministry of Agriculture and Rural Affairs in Zhuozhou City, China in October 2019. The capture size was divided into 4 classes, including 0, 1, 2, and 3 kg/m2, according to the quality of maize residues between rows. At the same time, each class included the front lighting on a sunny day, direct sunlight, backlight on a sunny and cloudy day. Moreover, the images of the 0 kg/m2 class also involved different shapes and surface moisture contents, due to the change of time and weather. The results revealed that the algorithm presented higher accuracy and robustness for the stubble row segmentation under various maize residues quality between rows and different lighting conditions. The average recognition accuracy of standing stubble was 93.8% in the SVM recognition model, whereas, those were 62.76% and 63.71% in the BPNN and ELM model, respectively. The average segmentation accuracy, average recall rate, and F1avr in the SVM recognition model were 93.72%，92.35% and 93.03%, respectively, whereas, those in the BPNN, ELM and genetic models were 61.88%, 86.94%, 72.3%, 62.92%, 88.75%, 73.63%, 90.13%, 51.36% and 65.43%, respectively. Additionally, the average processing time was 0.06 s for a 640×480(pixel) image using the SVM recognition models, indicating excellent real-time performance. Therefore, the SVM recognition model can widely be expected to realize better performance than others in the segmentation of the maize stubble row after the maize is harvested by the combine harvesters.