Bim Prasad Shrestha

Integrating reflectance and fluorescence imaging for apple disorder classification

Multispectral imaging in reflectance and fluorescence modes combined with neural network analysis was used to classify various types of apple disorder from three apple varieties (Honey Crisp, Red Cort, and Red Delicious). Eighteen images from a combination of filter sets and three different imaging modes (reflectance, visible light induced fluorescence, and UV induced fluorescence) were acquired for each apple sample as a base for pixel-level classification into normal or disorder tissue. Two classification schemes, a 2-class and a multiple class, were developed and tested in this study. In the 2-class scheme, pixels were categorized into normal or disorder tissue, whereas in the multiple class scheme, pixels were categorized into normal, bitter pit, black rot, decay, soft scald, and superficial scald tissues. Results indicate that single variety training under the 2-class scheme yielded highest accuracy with total accuracy of 95, 97, and 100 % for Honey Crisp, Red Cort, or Red Delicious respectively. In the multiple-class scheme, the classification accuracy of Honey Crisp apple for normal, bitter pit, black rot, decay, and soft scald tissue was 94, 93, 97, 97, and 94 % respectively. Through variable selection analysis, in the 2-class scheme, fluorescence models yielded higher total classification accuracy compared to reflection models. For Red Cort and Red Delicious, models with only FUV yield more than 95% classification accuracy, demonstrating a potential of fluorescence to detect superficial scald. Several important wavelengths, including 680, 740, 905 and 940 nm, were identified from the filter combination analysis. The results indicate the potential of this technique to accurately recognize different types of disorder on apple.

Shape feature extraction and pattern recognition of sand particles and their impact

Sand deposition is the major problem of Nepalese rivers and it causes substantial impact to different sectors including hydropower generation, natural resource management, and many others. Due to the typical nature of soil and sand of Nepalese mountains it has almost become impossible to predict and manage the upcoming natural disasters and hazards. Sand deposition in rivers affect landslides, aquatic life of rives, environmental disorders and many others. Sedimentation causes not only disasters but also reduces the overall efficiency of hydropower generation units as well. A systematic approach to the problem has been identified in this work. Sand particles are collected from the erosion sensitive power plants and its digital images have been acquired. Software has been developed on MATLAB 6.5 platform to extract the exact shape of sand particles collected. These shapes have further been analyzed by artificial neural network. This network has been first trained for the known input and known output. After that it is trained for unknown input and known output. Finally these networks can recognize any shape given to it and gives the shape which is nearest to the seven predefined shape. The software is trained for seven types of shapes with shape number 1 to 7 in increasing number of sharp edges. The shape with shape number seven is having large number of sharp edges and considered as most erosive where as shape with shape number one is having round edges and considered as least erosive.

Computational analysis of Pelton bucket tip erosion using digital image processing

Erosion of hydro turbine components through sand laden river is one of the biggest problems in Himalayas. Even with sediment trapping systems, complete removal of fine sediment from water is impossible and uneconomical; hence most of the turbine components in Himalayan Rivers are exposed to sand laden water and subject to erode. Pelton bucket which are being wildly used in different hydropower generation plant undergoes erosion on the continuous presence of sand particles in water. The subsequent erosion causes increase in splitter thickness, which is supposed to be theoretically zero. This increase in splitter thickness gives rise to back hitting of water followed by decrease in turbine efficiency. This paper describes the process of measurement of sharp edges like bucket tip using digital image processing. Image of each bucket is captured and allowed to run for 72 hours; sand concentration in water hitting the bucket is closely controlled and monitored. Later, the image of the test bucket is taken in the same condition. The process is repeated for 10 times. In this paper digital image processing which encompasses processes that performs image enhancement in both spatial and frequency domain. In addition, the processes that extract attributes from images, up to and including the measurement of splitters tip. Processing of image has been done in MATLAB 6.5 platform. The result shows that quantitative measurement of edge erosion of sharp edges could accurately be detected and the erosion profile could be generated using image processing technique.

Integrating Reflectance and Fluorescence Imaging for Apple Disorder Classification

Multispectral imaging in reflectance and fluorescence modes combined with neural network analysis was used to classify various types of apple disorder from three apple varieties (Honey Crisp, Red Cort, and Red Delicious). Eighteen images from a combination of filter sets and three different imaging modes (reflectance, visible light induced fluorescence, and UV induced fluorescence) were acquired for each apple sample as a base for pixel classification into normal or disorder tissue. Two classification models, a 2-class model and a 6-class model, were developed and tested in this study. In the 2-class model, pixels were categorized into normal or disorder tissue, whereas in the 6-class model, pixels were categorized into normal, bitter pit, black rot, decay, soft scald, and superficial scald tissues. Results indicate that single variety training under the 2-class model yielded highest accuracy with total accuracy of 95, 97, and 100 % for Honey Crisp, Red Cort, or Red Delicious respectively. In the 6-class model, the classification accuracy of Honey Crisp apple for normal, bitter pit, black rot, decay, and soft scald tissue was 94, 93, 97, 97, and 94 % respectively. The results indicate the potential of this technique to accurately recognize different types of disorder on apple.

Computational and experimental study of effects of sediment shape on erosion of hydraulic turbines

Hard particles as Quartz and Feldspar are present in large amount in most of the rivers across the Himalayan basins. In run-off-river hydro power plants these particles find way to turbine and cause its components to erode. Loss of turbine material due to the erosion and subsequent change in flow pattern induce several operational and maintenance problems in the power plants. Reduction in overall efficiency, vibrations and reduced life of turbine components are the major effects of sediment erosion of hydraulic turbines. Sediment erosion of hydraulic turbines is a complex phenomenon and depends upon several factors. One of the most influencing parameter is the characteristics of sediment particles. Quantity of sediment particles, which are harder than the turbine material, is one of the bases to indicate erosion potential of a particular site. Research findings have indicated that shape and size of the hard particles together with velocity of impact play a major role to decide the mode and rate of erosion in turbine components. It is not a common practice in Himalayan basins to conduct a detail study of sediment characteristics as a part of feasibility study for hydropower projects. Lack of scientifically verified procedures and guidelines to conduct the sediment analysis to estimate its erosion potential is one of the reasons to overlook this important part of feasibility study. Present study has been conducted by implementing computational tools to characterize the sediment particles with respect to their shape and size. Experimental studies have also been done to analyze the effects of different combinations of shape and size of hard particles on turbine material. Efforts have also been given to develop standard procedures to conduct similar study to compare erosion potential between different hydropower sites. Digital image processing software and sieve analyzer have been utilized to extract shape and size of sediment particles from the erosion sensitive power plants. The experimental studies of impact of different shapes and sizes of sediment particles on hydraulic turbine material have been conducted on two different test rigs method at Kathmandu University, High velocity test rig method and Rotating Disc apparatus (RDA) at Kathmandu University. Twenty one different sediment shape samples and four different sand size range were studied to correlate the effects of sediment shape and size with the erosion of turbine material. It was observed that the shape of sediment particles have considerable effect on erosion of turbine material. In general Irregular shapes have more erosion potential than regular shapes. It was also observed that the particles with the irregular shape of smaller size induce higher erosion rates than that of the larger size with the same shape. These findings will help to select the proper site of a power plant in erosion prone basins and would also help to design suitable settling basins to trap sediment particles having higher erosion potentials.

Quality Evaluation of Plant Products

Hyperspectral imaging or imaging spectrometry combines the strengths of computer vision technology with optical spectroscopy. It is primarily suited for measurement of parameters that vary spatially both at the external surface of samples and internally within the samples. The parameters may be physical features such as incipient bruises or surface contamination, or chemical constituents such as sugar and acidity. While the acquisition of images generally follows the procedures of machine vision, adding a spectral dimension would require the rigor of multivariate statistics, also known as chemometrics, to find functional relationships between the measured values and target parameters. Its application to agriculture, particularly to post-harvest processing, has recently been explored by university research laboratories in order to develop new techniques for non-destructive measurement of quality.

Opto-electronic determination of insect presence in fruit

Opto-electronic methods represent a potential to identify the presence of insect activities on or within agricultural commodities. Such measurements may detect actual insect presence or indirect secondary changes in the product resulting from past or present insect activities. Preliminary imaging studies have demonstrated some unique spectral characteristics of insect larvae on cherries. A detailed study on spectral characteristics of healthy and infested tart cherry tissue with and without larvae (Plum Curculio) was conducted for reflectance, transmittance and interactance modes for each of UV and visible/NIR light sources. The intensity of transmitted UV signals through the tart cherry was found to be weak; however, the spectral properties of UV light in reflectance mode has revealed some typical characteristics of larvae on healthy and infested tissue. The larvae on tissue were found to exhibit UV induced fluorescence signals in the range of 400-700 nm. Multi spectral imaging of the halved tart cherry has also corroborated this particular behavior of plum curculio larvae. The gray scale subtraction between corresponding pixels in these multi-spectral images has helped to locate the larvae precisely on the tart cherry tissue background, which otherwise was inseparable. The spectral characteristics of visible/NIR energy in transmittance and reflectance mode are capable of estimating the secondary effect of infestation in tart cherry tissue. The study has shown the shifting in peaks of reflected and transmitted signals from healthy and infested tissues and coincides with the concept of browning of tissue at cell level as a process of infestation. Interactance study has been carried out to study the possibility of coupling opto-electronic devices with the existing pitting process. The shifting of peaks has been observed for the normalized intensity of healthy and infested tissues. The study has been able to establish the inherent spectral characteristic of these tissues. It was found that there existed promising futuristic possibilities to use opto-electronic sensing to estimate the degree of secondary effect of insect activities within the tissue.

Basic Study on Strawberry Harvesting Robot (Part II)

Abstract These papers describe a strawberry harvesting robot where by robot vision and a 4 DOF’s Cartesian coordinate manipulator were used. This first part presents the development of a robot vision system and the algorithm for locating and feature extracting of strawberry fruits. The robot vision system employs the use of two color CCD cameras. The first camera is used to capture the whole area image under focus within the harvesting range, and the second camera captures only the image of the strawberry fruit to be plucked. The algorithm converts the captured images from RGB to L*a*b and extracts recognized position, orientation and shape of strawberry from a gray image of the L*a*b color model. Experimental results show that the robot vision system can extract position, orientation and shape of various strawberries in ordinary lighting condition. The Part II presents the design and development of the robot frame, the plucking hand, 4 DOF’s Cartesian coordinate manipulator and the Control System.

Classification of biological and non-biological fluvial particles using image processing and artificial neural network

Particles flowing along with water largely affect safe drinking water, irrigation, aquatic life preservation and hydropower generation. This research describes activities that lead to development of fluvial particle characterization that includes detection of biological and non-biological particles and shape characterization using Image Processing and Artificial Neural Network (ANN). Fluvial particles are characterized based on multi spectral images processing using ANN. Images of wavelength of 630nm and 670nm are taken as most distinctive characterizing properties of biological and non-biological particles found in Bagmati River of Nepal. The samples were collected at pre-monsoon, monsoon and post-monsoon seasons. Random samples were selected and multi spectral images are processed using MATLAB 6.5. Thirty matrices were built from each sample. The obtained data of 42 rows and 60columns were taken as input training with an output matrix of 42 rows and 2 columns. Neural Network of Perceptron model was created using a transfer function. The system was first validated and later on tested at 18 different strategic locations of Bagmati River of Kathmandu Valley, Nepal. This network classified biological and non biological particles. Development of new non-destructive technique to characterize biological and non-biological particles from fluvial sample in a real time has a significance breakthrough. This applied research method and outcome is an attractive model for real time monitoring of particles and has many applications that can throw a significant outlet to many researches and for effective utilization of water resources. It opened a new horizon of opportunities for basic and applied research at Kathmandu University in Nepal.

Opto-Electronic determination of insect presence in fruit

Opto-electronic methods represent a potential to identify the presence of insect activities on or within agricultural commodities. Such measurements may detect actual insect presence or indirect secondary changes in the product resulting from past or present insect activities. Studies have progressed from preliminary multispectral and fluorescence imaging, to more detailed tissue spectral measurement, analysis and classification of infested and healthy whole and tissue samples of cherries and also the development and evaluation of a prototype sensor probe. Infested whole cherries were detected with complete spectra information with accuracy as high as 80-85%. A prototype dedicated probe utilizing two LED-based wavebands of incident light performed well but somewhat below the complete spectra accuracy. Additional analysis for optimal wavelengths demonstrated the potential to increase total classification to near 90%.