Abstract: Abstract: Impurity separation test is widely used to detect the impurity rate of seed cotton in purchasing, due mainly to the fact that some foreign matter can be picked manually. However, the average consuming time for a cotton sample test is 20-30 minutes. A new testing procedure is, therefore, necessary to rapidly detect the impurities rate, and thereby to classify the seed cotton for high production efficiency. In this study, an intelligent measurement system was developed using machine vision to detect the impurity rate of machine-picked seed cotton in acquisition. The whole system consisted of a drive transmission unit, a cotton pressing unit, a sensor unit, a machine vision system, and a control system. A cleaning machine for large impurities was utilized to mechanically remove the cotton stems and hulls from the 500 g sample of seed cotton. The cotton sample without large impurities was weighed, and then transported into the machine vision system. The surface of seed cotton was automatically compacted by the pressing unit, aiming to reduce the influences of uneven brightness and shadows in subsequent image acquisition. The RGB double-sided imaging was selected to acquire the image of seed cotton. The homomorphic filtering and Principal Component Analysis (PCA) were selected to preprocess the collected RGB images. A local adaptive threshold was then utilized to segment the preprocessed images into the impurities and cotton. After that, the segmented regions of impurities were calculated to predict the weight of small impurities. The total weight of impurities was equal to the predicted value of small impurities and the measured value of large impurities. The rate of small impurities was the ratio of predicted value to the sample weight without large impurities. The final impurity rate was achieved for the total weight of impurities in the total 500 g sample of seed cotton. Linear Regression (LR) and support vector regression (SVR) models were used to compare the predicted accuracy. The LR achieved a better performance, where the determination coefficient R2 for the final impurities rate was 0.95, and the root-mean-square error was 0.58%. The mean absolute error of the final impurity rate was 0.36 percentage points of 100 extra testing samples. The processing time was 48.38 s for a single detection of sample impurity rate. The findings can provide a sound reference to the measuring equipment for the impurity rate of machine-picked seed cotton during the acquisition process.