Advanced Instrumentation for Detection of Defects and Diseases in Sericulture

Abstract

Sericulture is the art of rearing silkworms for silk production. The matured silkworms will be spinning cocoons as a protective shield before they undergo metamorphosis. These cocoons are immersed in boiling water to soften, and raw silk yarn is reeled from them. Being silkworms are sensitive to environmental changes, they are susceptible to many diseases. The diseased and weak silkworms spin defective cocoons, which should be removed from the production of good quality silk. In developing countries, the diagnosis of diseases and defects depends mainly on subjective and unreliable methods due to the unaffordability of customized hi-fidelity support systems. Any erroneous findings will cause immense loss to the farmers and reelers. This thesis reports on advanced yet affordable instrumentation systems to detect a) the external defects and b) internal defects in cocoons, and c) pebrine in silkworms.

a) The proposed system for detecting cocoons' external defects consists of an image acquisition and processing unit made using a smart-camera. The LED strip and cardboard box are used for conditioned illumination. The cocoons are rolled over a slope, and their whole surfaces are imaged and processed utilizing the smart-camera. An image processing algorithm, exploiting the techniques such as morphological operations, image enhancement and ellipse fitting, is developed for quantitative measurements of the cocoon's size, shape and colour. These values are compared against each category's thresholds, and identified the cocoon category (good, externally stained, double, urinated and uzi pierced). A graphical user interface is designed for helping farmers and reelers to visualize the quality of cocoons easily. The developed system is tested on 137 cocoons, and the results showed that it could evaluate 96 cocoons per second with 100% accuracy.

b) The instrument developed for the automated detection of cocoons' internal defects is designed to have a plastic arm attached to an Arduino-driven servo motor to vibrate the cocoons. Two microphones with specific separation are used to record the vibration induced acoustic emission (VIAE) from cocoons and ambient noise. A spectral analysis algorithm is developed to identify the cocoon category (good, dried and mute) based on the area under VIAE's power spectral density curve. This system classified 86 cocoons with 100% accuracy at a speed of 20 seconds/cocoon. These two defect detection systems are compatible to be integrated for the detection of external and internal defects of all cocoons in a given lot.

c) In the developed pebrine diagnostic instrument, a machine learning algorithm classifies the quantitative phase images of test spores in silkworm's body fluid as pebrine and Metarhizium anisopliae (MA), which resembles pebrine spore in microscopic examination. Its hardware consists of a custom-made motorized brightfield microscope to acquire a pair of (focused and defocused) images of spores. A transport of intensity equation based algorithm is developed to produce phase images, which carry information of spores' internal features. Further, the histogram of oriented gradients (HOG) feature of the phase images is extracted and used to train a machine learning algorithm. This system is tested on 92 pebrine and 185 MA spores and found to offer a classification accuracy of 97%. This instrument can be used for diagnosing other silkworm diseases such as nuclear polyhedrosis and septicemia.