NEURAL NETWORK CLASSIFICATION OF WEEDS VERSUS CROPS USING MULTISPECTRAL INFORMATION
Network Classification of Weeds versus Crops using Multispectral imformation. Effective weed control is a critical issue in agriculture. Weed infestations reduce yields by competing with crops for nutrients. For this reason farmers spend a great deal of money and effort in controlling weeds. American farmers alone annually spend $3 billion in chemicals and $1 billion in application technologies to protect $16 billion in crops. Health care and environmental costs associated with the wide spread use of these chemicals is estimated to be in excess of $8 billion annually in the U.S. alone In this work multispectral reflectance images are used in conjunction with a neural network classifier for the purpose of detecting and classifying weeds under real field conditions. Multispectral reflectance images which contained different combinations of weeds and crops were taken under actual field conditions. This multispectral reflectance information was used to develop algorithms that could segment the plants from the background as well as classify them into weeds or crops. In order to segment the plants from the background the mult ispectral reflectance of plant 5; and background were studied and a relationship was derived. It was found that using a ratio of two wavelength reflectance images (750nm and 670nm) it was possible to segment the plants from the background. Once this was accomplished it was then possible to classify. the segmented images into weed or crop by use of the neural network. The neural network developed for this work is a modification of the standard learning vector quantization algorithm. This neural network was modified by replacing the timevarying adaptation gain with a constant adaptation gain and a bina,ry reinforcement function. This improved accuracy and training time as well as introducing several new properties such as hill climbing and momentum addition. The network was trained andl tested with different wavelength combinations in order to find the best results. Finially,t he results of the classifier were evaluated using a pixel based method and a block based method. In the pixel based method every single pix'el is evaluated to test whether it was classified correctly or not and the best weed classification results were 81.15% and its associated crop classification accuracy is 57.74%. In the block based classification method, the image was divided into blocks and each block was eva~luated to determine whether they contained weeds or not. Diflferent block sizes anti thresholds were tested. The best results for this method were 97.10% for a block size of 8 inches and a pixel threshold of 60.
Date of this Version