This paper focuses on the development of an intelligent machine vision for detection of field-fungi affected wheat kernels, where more than half of the seed coat is discolored. Presence of surface-defected (discolored) kernels in wheat seeds affects their quality, which consequently leads to substantial reduction in crop-yield. In order to determine quality estimation of wheat seeds, amount of surface-defected kernels in the same is required to be ascertained accurately. The present paper describes the development of a machine vision system for detection of surface-defected kernels in wheat seeds. In manual visual inspection, presence of surface-defected wheat kernels in a given sample is decided on the basis of the surface colour and texture attributes of kernels. However, in computer environment, both these attributes are quite difficult to quantify. In the present work, solution to this problem is provided by making use of new type of regional surface descriptors of wheat kernels using colour and texture attributes. Based on the quantitative description of these attributes, a neural classifier is executed to estimate quality of wheat seeds. In an attempt to achieve faster convergence, proposed neural classifier is trained with Levenberg-Marquardt (LM) learning algorithm. The results of present investigations indicate an average accuracy of more than 98.8% for the proposed system. The results of these investigations are quite convincing.

Login to Download PDF

Please login with your Paper ID and password to access the full PDF.

🔑 Login to Download