DEVELOPMENT AND PERFORMANCE TEST OF BANANA RIFE DETECTION TOOL USING LASER LIGHT BACKSCATTERING IMAGING

• Main Author: Intan Kuala, Seri
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/70161
• Institution: Institut Teknologi Bandung

Nondestructive Test (NDT) in postharvest technology has developed rapidly, but it is still quite expensive, complicated, and has low accuracy. Laser backscattering imaging is one of the optical-based NDT technologies that has a system that is easier to assemble, and customize and has high accuracy. In this research, a system based on the backscattering of laser light on Cavendish bananas has been designed with a mounting frame with dimensions of 80 x 80 x 80 cm, the laser as a light source with a wavelength of 650 nm 5 mW is placed on the top of the mount frame, forming an angle of + 15o with the sample surface. A detector in the form of a DSLR camera with a focal length of 18–105 mm placed close to the laser will record the laser beam backscattering image. A PC equipped with Digicam Control and Fiji ImageJ software serves to store and process the laser beam backscatter captured image. Observation of the backscattering parameters of laser light is the shape of the reflection and scattering of light by the banana surface, such as area, major, and minor axes, while from statistical processing, the light intensity is the intensity of max, min, mean, median, and mode. On the other hand, measurements were carried out using standard equipment including color measurements (L*, a*, b*) using the NH300 Portable Colorimeter, modulus of elasticity (ME) using the TA.XT plus C Texture Analyzer, as well as chemical parameters brix and pH using the ATAGO pocket refractometer PAL-1 with 0.1% resolution, 0.2% accuracy, and pH Meter Lab 855. These measurements were used to compare the results of the statistical analysis of laser backscattering on banana peels. Observations were made at 6 (six) points on both sides of the banana, except for chemical parameters (whole bananas), with a total of 5 replicates. Samples of 90 (ninety) Cavendish banana fingers (Musa acuminata colla) had the same biological and physiological age, were free from defects, and were stored at +17 oC. The results of the initial data processing showed that the pattern of changes in the middle part of the banana was the most consistent in all parameters, so that further data processing only used this part. The overall correlation value between laser backscattering parameters with physical, chemical, and ripeness is 96.875%, having a perfect relationship, 43.75% having a very strong relationship, and 62.5% having a strong relationship. From this correlation, a regression equation model can be formed, and then the performance of the system that has been created can be measured. The highest performance at b* is 97.55%, and the maturity level is 87.63%. The best performance results are achieved through machine learning (ML) algorithms on k-Nearest Neighbor (kNN) with 99.1% accuracy. kNN was able to correctly predict >=79.5% at 6 (six) maturity levels and >=93.5% at 2 (two) maturity levels. These results are quite good as the basis for this system to be able to become an alternative to the NDT method to detect banana ripeness levels that are more accurate, easy to assemble, and low cost.