Recognition of greenhouse cucumber fruit using computer vision

A method for cucumber fruit recognition in greenhouses is presented for a robotic fruit picking system. A three-layer back propagation (BP) neural network was set up to segment the cucumber plant images. The B (blue) and S (saturation) components were extracted as the input of the network. The multiple colour space feature fusion reduced illumination effects and enhanced image colour information. After successful training of the network, the cucumber plant images were segmented. The fruiting regions were preserved while most other portions were removed. Then, a binary image was made after morphologic processing and fruits were recognised by a logical operation with two templates based on the discriminated image. Finally, by a texture analysis a "third moment" was selected as a feature to identify the upper part of the fruit, which provides grip position for a robot endeffector. The experimental results on 40 cucumber plant images show that the recognition rate of fruits is about 76%.