Pulse coupled neural network (PCNN) and mathematical morphological technologies were employed to separate the mature in-greenhouse cucumber from complex background image. Four geometric feature values and three texture feature values based on gray level co-occurrence matrix (GLCM) of every connected regions in image were extracted, which were the input feature vector of least squares support vector machine (LS—SVM). The trained classifier was used for identifying the cucumber in image. Experimental results showed that 70 cucumber images were used for testing, the average rate of correct identification reach to 82.9% in different conditions, indicating that the method based on PCNN and LS—SVM could be used for in-greenhouse cucumber recognition.