dc.description.abstract |
Wheat is a crop of great importance, and as such, the market regulations governing its circulation
must be followed. The quality of wheat samples is now checked manually by human experts using
visual evaluation, and the contents are categorized into foreign matter, rotten and diseased, healthy,
broken, discolored, shriveled, and pest damaged kernels. Visual analysis, however, calls for a large
investment of time as well as skilled personnel. Furthermore, bias and inconsistent behavior
inherent in human nature have an impact on it. In the absence of complete au tomation, this
approach will not be adequate for large-scale inspection and grading.
The objective of this research is to develop a model that can evaluate the quality of wheat sample
constituents using digital image processing and Transfer Learning Techniques based on the
standard for wheat set by the Ethiopian Standards Agency. In order to segment and provide the
groundwork for feature extraction, a unique segmentation technique is proposed. To model the
components of a wheat sample, a total of 22 features have been identified.
A Vgg16-CNN network classifier using a backpropagation learning algorithm (with one hidden
layer, three dense layer and one dropout layer) has been developed to classify wheat samples. It
includes 22 input nodes and 7 output nodes, which correspond to the number of features and
classes, respectively. The network is trained, and its effectiveness is assessed against that of other
classifiers using both empirical data and evidence from the literature. A total of 2,889 kernels and
foreign objects have been gathered in order to train the classifier. 70% of the data is divided at
random between training and testing (30%). Overall classification accuracy for the classifier was
97.90 %. The accuracy rates for foreign, rotten and diseased, healthy, broken, discolored,
shriveled, and pest-damaged kernels are 100%, 95.30%, 98.70%, 98.10%, 100%, 98.50%, and
95.20%, respectively.
Keywords: Wheat quality assessment, Image Processing, Feature extraction, Segmentation, Classification |
en_US |