Application Of Multi-Layer Perceptron Technique To Detect And Locate The Base Of A Young Corn Plant

Vision based techniques have been widely used in precision farming especially to control the application of chemical products on a specific area. This can help minimizing the risk of soil and water pollution due to excessive application of chemical products. Machine vision can be used to gather information while the vehicle pulling the herbicide sprayer is in motion. This information can be processed, analyzed, and transformed into inputs for a decisional algorithm that controls the sprayer nozzle action in real-time. In this research, a vision system algorithm has been developed to identify and locate base of young corn trees based upon robot vision technology, pattern recognition techniques, and knowledge-based decision theory. Results of studying color segmentation using machine-learning algorithm and color space analysis is presented in this thesis. RGB (red, green, blue) color space data points on an image are projected into HSV (hue, saturation, value) color space to provide data points that are insensitive to the variations of illumination in outdoor environment. Multi-layer perceptron (MLP) neural network trained using backpropagation algorithm is used to segment the color image. The results of color segmentation show that the algorithm is able to segment the images reliably with less appearance of small blobs. Morphological operation is applied to remove the small blobs. Prior to localization of the base of young corn tree, skeletonizing operation is performed to get the basic shape of the object. Another structure of MLP trained using backpropagation algorithm is used to detect and locate the base of the young corn tree using the skeleton of the segmented image. Prior to choosing the MLP structures for both color segmentation and object detection, a number of experiments have been conducted to find the best MLP structures that can give considerably good recognition and classification rate with considerable amount of processing time required. Results of the experiments to find the best MLP structures are presented together with the discussion. The recognition rate is presented and compared with another related research work, where the results show equal performance of both algorithms. This shows that machine-learning algorithm such as MLP is a viable method for color segmentation as well as object recognition.

Preview PDF FK_2007_43.pdf Download (451kB)

Actions (login required)

View Item