The speedy evolution of population as well as global warming are giving rise to different challenges to the farmers. Together with the effort to raise the farming production, profit maximization al so need to be focused by them. The principal phase in self-ruling crop management is selective weed treatment. Trustworthy and precise detection of weeds to lessen the destruction of neighboring plants is a crucial task to the farmers however. Convolutional Neural Network (CNN) cascaded with an encoder-decoder for weed classification is devised by the current work. But still the noise existing in the image is not eliminated. Thus, a less significant classification operation is realized. An Adaptive Median Filtering (AMF) with Modified Convolutional Neural Network (MCNN) is therefore devised by the proposed system to resolve this issue in order to enhance the accurateness of classification. Near-Infrared (NIR) and red images are considered as input image sat the start. From NIR and red patch images, so as to obtain Normalized Difference Vegetation Index (NDVI), the fundamental automated image processing methods are applied. To effectively take away the noise from images, then the Adaptive Median Filtering (AMF) algorithm is brought in. By means of Quad Histogram, the extraction of color feature is accomplished then. Upon the images, the quad tree decomposition is operated and homogenous blocks with various sizes are indicated. Well-defined edges amongst vegetation and others is achieved by an auto-threshold boundary detection technique and for selection of threshold on resultant image, Otsu’s method is applied. To categorize the images into crop, weed and background, Modified Convolutional Neural Network (MCNN) is employed as a final step. Thus, with respect to recall, f-measure and precision, proposed system realizes superior performance in comparison with the existing system as shown by the experimental results.
Volume 12 | 04-Special Issue
Pages: 85-96
DOI: 10.5373/JARDCS/V12SP4/20201469