Abhra Pal

X-Ray Imaging and General Regression Neural Network (GRNN) for Estimation of Silk Content in Cocoons

This paper proposes a non-destructive technique for silk content estimation in cocoons. The price of a cocoon is determined by the silk content which is determined manually by visual inspection or feeling the toughness of the cocoon shell. The above methods are subjective, non-repeatable and prone to human error. With such non-transparent conventional methods of silk estimation, the buyers and sellers are unhappy over any transaction. Our proposed non-destructive technique uses soft x-ray image analysis technique backed up by soft computing algorithm to estimate silk content. Advance image processing and analysis techniques have been applied to extract morphological features from the x-ray images of the cocoons and features are fed to GRNN to estimate the silk content. Total 594 tasar cocoons have been analyzed with the developed solution and the results have been validated with human experts. Accuracy of the system for silk content estimation has been calculated as more than 85%.

Quality inspection of cocoons using X-ray imaging technique

Quality of cocoons determined by its silk content which is directly related with the market value. Silk is broadly used in textiles and it is costly too. Farmers are selling the cocoons in a bulk with an average market value to the yarn producers. Presently, the quality of cocoons is verified by visualizing the color, size, shape and feeling its solidity on pressing by fingers of our hand. These manual inspections sometimes create dissatisfaction among the buyers and sellers. Sometimes, the sellers (farmers) are duped by the clever buyers. In other method, the silk content is estimated by taking the average raw cocoon shell weight after cutting the cocoon and removing the pupa from it. This approach is destructive, time consuming, expensive and laborious also. In this paper, X-ray imaging technique has been explored to estimate the silk content and determine the quality of cocoons. Firstly the images of the cocoons are captured using standard X-Ray imaging setup. Then the images are enhanced using digital image processing techniques. Finally, different dimensional features are extracted using image analysis techniques. A new method for estimation of the silk content has been proposed using the GRNN (General Regression Neural Network). Quality of the cocoons has been evaluated using unsupervised artificial neural network technique known as SOM (Self Organizing Map) which produces the different classes of quality grades of cocoons. In this experiment, we have considered five classes - good, medium, bad, dead pupa and un-identified quality. Total 49 no of cocoons have been used for the experimentation. The result shows that using GRNN the estimation of silk content is quite helpful with a fair level of accuracy. Using SOM technique, quality of cocoons has been determined and the result is validated with the manual inspection method. Both this approach of estimating the silk content and determining the quality of cocoons opens new possibilities in the field of automatic, non-destructive technique for price appraisal of cocoons.

Development of machine vision solution for grading of Tasar silk yarn

Quality of Tasar fabric demands uniform coloured silk yarn during weaving. But, the variation of yarn colour depends on various natural factors like eco-race and feeding of silk worms, weather conditions etc and other production factors. So, silk yarns need to be sorted after production. At present, yarns are sorted manually by a group of experts which is subjective in nature. Again, due to lustrous nature of silk yarn, it reflects light and therefore it is difficult to ascertain the exact colour manually. Slight variation in colour is difficult to detect manually but the market demands lots with perfectly uniformly coloured yarns within the lot though the inter-lot variation in colour is encouraged. So, there is need to develop a solution which can grade the silk yarn objectively, reliably and mimic the human perception. This paper proposes a new machine vision solution for automatic grading of silk yarn based on its colour. The system consists of an enclosed cabinet which encompasses of a low cost digital camera, uniform illumination arrangement, weighing module, mechanical arrangement for sample holding and a grading software which applies image analysis technique using CIELab colour model with rotational invariant statistical feature based hierarchical grading algorithm for colour characterization. Performance of the system has been validated with the human experts and accuracy has been calculated as 91%.

Development of photomicrographic image analysis solution for sporozoa detection in Tasar moth

In spite of the availability of natural resources and traditional skills, Tasar sericulture in India is stagnating due to frequent outbreak of a number of diseases. The most common and deadliest among all is Pebrine disease caused by a microsporidian parasite Nosema sp. Infections of the disease range from chronic to highly virulent and can result in complete loss of crop. The disease has become increasingly more and more complex as more number of microsporidian strains infecting silkworms is being identified. Therapeutic methods to control the disease at commercial scale have so far been proven to be ineffective. As of now, preventive methods are generally followed to restrict the disease below the danger threshold. As the disease is trans-ovarially transmitted hence the common method is to eliminate primary infection at the egg stage by testing the body fluid of the egg laying moths under microscope. If the tissues are found free of infection, then only the corresponding eggs are distributed amongst the villagers pursuing sericulture. Currently the entire process is manual, time and labour intensive. Many a time human error also creeps in leading to outbreak of the disease. This paper proposes automation of the disease detection process by capturing photo-micrographic images and classifying spores using digital image analysis technique thereby improving productivity and accuracy of this process. The proposed solution has been tested in the tasar grainages and the software results have been validated with the human experts. The accuracy of correct identification of Pebrine spores has been found as 87%.

Disease detection in Tasar moth using micrographic image analysis solution

Tasar Silk is one of the most widely used luxurious material not only in India but in all over the world. Tasar cocoons are used as the raw material for silk production. In sericulture industry one of the most important perspectives is disease free egg production for formation of healthy cocoons. Out of the many different diseases, Pebrine is the most destructive one which spreads transovarially and makes huge loss in Tasar silk industry. As of now, the egg laying moths are inspected manually by field experts using a low cost student microscope with 675X magnification in day light. Problems with the present methodology are of many folds - lack of field experts in remote places, tedious and time consuming process in case of visual inspection, insufficient light in cloudy atmosphere, lack of authenticity due to manual inspection etc. In this paper we have proposed a motorized microscopic image analysis solution which overcomes the above shortcomings. Positioning of the slide has been controlled using 2 nos of stepper motors and focusing of microscope has been done by introducing another stepper motor. Slides are prepared by taking smear of mother moth from its lower part of the abdomen and mixing with medium concentration (0.5 gm/ 100 ml water) of K2CO3 solution. Slides are placed manually under the microscope and images are captured using a low cost digital camera placed on the eyepiece of the microscope. Image analysis software has been developed to identify the Pebrine using different morphological features. More than 200 images have been analyzed using developed solution & the results have been validated with the human experts. Accuracy of the Pebrine disease detection solution in case of Tasar moth has been calculated as 85%.

Estimation of Theaflavins (TF) and Thearubigins (TR) Ratio in Black Tea Liquor Using Electronic Vision System

Quality of black tea is generally assessed using organoleptic tests by professional tea tasters. They determine the quality of black tea based on its appearance (in dry condition and during liquor formation), aroma and taste. Variation in the above parameters is actually contributed by a number of chemical compounds like, Theaflavins (TF), Thearubigins (TR), Caffeine, Linalool, Geraniol etc. Among the above, TF and TR are the most important chemical compounds, which actually contribute to the formation of taste, colour and brightness in tea liquor. Estimation of TF and TR in black tea is generally done using a spectrophotometer instrument. But, the analysis technique undergoes a rigorous and time consuming effort for sample preparation; also the operation of costly spectrophotometer requires expert manpower. To overcome above problems an Electronic Vision System based on digital image processing technique has been developed. The system is faster, low cost, repeatable and can estimate the amount of TF and TR ratio for black tea liquor with accuracy. The data analysis is done using Principal Component Analysis (PCA), Multiple Linear Regression (MLR) and Multiple Discriminate Analysis (MDA). A correlation has been established between colour of tea liquor images and TF, TR ratio. This paper describes the newly developed E‐Vision system, experimental methods, data analysis algorithms and finally, the performance of the E‐Vision System as compared to the results of traditional spectrophotometer.