SOFTWARE DEVELOPMENT FOR BLACK TEA'S PHYSICAL VARIABLE AND QUALITY CLASS RELATIONSHIP ANALYZING

• Main Author: PRASTA JENIE (NIM 23506011), RENAN
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/9045
• Institution: Institut Teknologi Bandung

Black tea quality evaluation is a standardized quality assurance method which is held by every tea plantation. Visual assertion is held to measure tea.s visual quality parameter such as shape, size, colour, tips, texture, and uniformity. Everyone perception can be different, so effort to build such an unbiased system is needed. To cope with, it is suggested to let a tool that can extract the tea quality parameters without personal bias do the job. The tools suggested based on image processing and artificial neural network. Image processing methods are used to get the measureable parameters from the black tea.s particle, and artificial neural network is used to infer the quality parameter based on it. The development of software that based on image processing and artificial neural network demanded several actions. <p> <br /> <br /> <br /> <br /> <br /> 1. How to measure black tea.s visual quality parameter such as geometry, colour, and its derivation. <p> <br /> <br /> <br /> <br /> <br /> 2. How to reassure image processing.s consistency. <p> <br /> <br /> <br /> <br /> <br /> 3. How to extract the black tea.s quality parameter as the measurable parameter <p> <br /> <br /> <br /> <br /> <br /> 4. How to choose and reassure performance of methods or artificial neural network used. <p> <br /> <br /> <br /> <br /> <br /> 5. How to reassure developed device performance <p> <br /> <br /> <br /> <br /> <br /> The artificial neural network is chosen by filtering the features needed for evaluating black tea out of overall features of choosable artificial neural networks. Disregarding excessive features, the 6 usable artificial neural networks is back propagation network, radial function network, Jordan network, recurrent network with mental model, spiking neural network, and cascading neural network. It is proposed to separate development in two phases. First phase is to measure system performance, and second phase software is build based on the first software. <p> <br /> <br /> <br /> <br /> <br /> 1. First phase software is used to get as many measurable parameter as possible from the object and picking the best possible parameter combination based on its correlation to tea quality parameter. <p> <br /> <br /> <br /> <br /> <br /> 2. Second phase software is build upon first and aimed toward system performance. <p> <br /> <br /> <br /> <br /> <br /> The software built to cope with this arrangement of task. <p> <br /> <br /> <br /> <br /> <br /> 1. Extraction of black tea.s physical parameters. The tool work to distinguish every non background object, and extracting the physical parameters out of them. <p> <br /> <br /> <br /> <br /> <br /> 2. Calculation of correlation between black tea.s measurable and quality parameters. <p> <br /> <br /> <br /> <br /> <br /> 3. The user can choose the desirable parameter for the artificial neural network used. <p> <br /> <br /> <br /> <br /> <br /> 4. The artificial neural network is usable in real condition test. <p> <br /> <br /> <br /> <br /> <br /> System performance testing is held using real life data. <p> <br /> <br /> <br /> <br /> <br /> 1. Testing is done using 2 sample mode, stacked and scattered, 3 artificial neural network structure, 3-5-1, 11-21-1, and 11-21-5, and 3 input parameters structure, standard, correlation optimalized, and correlation disoptimalized, arranged into 18 test cases. Testing done using real life samples not optimalized samples. <p> <br /> <br /> <br /> <br /> <br /> 2. From black tea.s parameters correlation data, Mean RGB-IG is one of the best for stacked samples, and Area / mean HSL-S is one of the best for scattered samples. <p> <br /> <br /> <br /> <br /> <br /> 3. The test shows that maximizing correlation can yield better training speed across every test cases, according preceding arrangement, smaller better, [0.016, 0.013, 0.018], [0.003, 0.003, 0.048], [0.035, 0.020, 0.040], [0.013, 0.012, 0.014], [0.012, 0.009, 0.0012], and [0.039, 0.031, 0.037]. <p> <br /> <br /> <br /> <br /> <br /> 4. The test shows that maximizing correlation can yield better testing accuracy across 50 % of test cases, according preceding arrangement, bigger better, [0.87, 0.56, 0.20], [0.73, 0.77, 0.20], [0.75, 0.72, 0.76], [0.26, 0.22, 0.21], [0.23, 0.27, 0.18], and [0.31, 0.39, 0.34]. <p> <br /> <br /> <br /> <br /> <br /> 5. The scattered black tea.s sample measurable parameters correlation to quality parameter is not high enough to give better accuracy. <p> <br /> <br /> <br /> <br /> <br /> This research only had done the black tea.s particle visual and colour part. Further research is needed in black tea.s particle aroma, liquor colour and aroma, and resides colour. Other research needed is in the device shape of the black tea.s automation system.