A new method along with Zernike moments for classify fruit shape is developed, the image is first subjected to a
normalization process using its regular moments to obtain scale and translation invariance, the rotation invariant Zernike
features are then extracted from the scale and translation normalized images and the numbers of features are decided by
primary component analysis (PCA), at last, these features are input to support vector machine (SVM) classifier. This
method performs better than traditional approaches because of their orthogonal base and rotation invariance of the
defined features on them, which is verified by experiments on Zernike moments and Fourier descriptors.
Applications of machine vision in automated inspection and sorting of fruits have been widely studied by scientists and. Preprocess of the fruit image is needed when it contain much noise. There are many methods for image denoise in literatures and can acquire some nice results, but which will be selected from these methods is a trouble problem. In this research, total variation (TV) and shock filter with diffusion function were introduced, and together with other 6 common used denoise method s for different type noise type were tested. The result demonstrated that when the noise type was Gaussian or random, and SNR of original image was over 8,TV method can achieve the best resume result, when the SNR of original image was under 8, Winner filter can get the best resume result; when the noise type was salt pepper, median filter can achieve the best resume result
A new method along with shape descriptor using support vector machine for classify fruit shape is developed, the image is
first subjected to a normalization process using its regular moments to obtain scale and translation invariance, the rotation invariant
Zernike features are then extracted from the scale and translation normalized images and the numbers of features are decided by
primary component analysis (PCA), at last, these features are input to support vector machine (SVM) classifier and are compared to
different classifiers. This method using support vector machine as classifier performs better than traditional approaches that is
verified by some experiments.
Fourier transform near infrared reflectance (FT-NIR) spectroscopy has been used successfully to measure soluble
solids content (SSC) in citrus fruit. However, for practical implementation, the technique needs to be able to compensate
for fruit temperature fluctuations, as it was observed that the sample temperature affects the near infrared reflectance
spectrum in a non-linear way. Temperature fluctuations may occur in practice because of varying weather conditions or
improper conditioning of the fruit immediately after harvest. Two techniques were found well suited to control the
accuracy of the calibration models for soluble solids with respect to temperature fluctuations. The first, and most
practical one, consisted of developing a global robust calibration model to cover the temperature range expected in the
future. The second method involved the development of a range of temperature dedicated calibration models. The
drawback of the latter approach is that the required data collection is very large. The global temperature calibration
model avoids temperature-sensitive wavelengths for the calibration of SSC. Global temperature models are preferred
above dedicated temperature models because of the following shortcomings of the latter. For each temperature, a new
calibration model has to be made, which is time-consuming.
The Gradient Vector Flow (GVF) snake was used for color fruit shape detection, which is proposed by Chengxiang
Xu, this snake has two well properties than traditional snake: large capture range and its ability to move into
boundary concavities. Indicator and morphological operation before applying GVF snake firstly preprocess the color
fruit image. In our experiments, we compared the detection result of this approach to traditional snake and traditional
edge operators and it is obvious that the performance of this approach is better; the boundaries detected by GVF
snake are thin and smooth, which are very important for fruit size detection and shape classification.
In this research, a new algorithm for fruit shape classification was proposed. The level set representations according to signed distance transforms were used, which are a simple, robust, rich and efficient way to represent shapes. Based on these representations, the rigid transform was adopted to align shapes within the same class, and the simplest possible criterion, the sum of square differences was considered. After align procedure, the average shape representations can easily be derived and shape classification was performed by the nearest neighbor method. Promising results were obtained on experiments showing the efficiency and accurate of our algorithm.
Key words: machine vision, shape classification, fruit sorting, level set
Shape is one of the major concerns and which is still a difficult problem in automated inspection and sorting of fruits. In this research, we proposed the multi-scale energy distribution (MSED) for object shape description, the relationship between objects shape and its boundary energy distribution at multi-scale was explored for shape extraction. MSED offers not only the mainly energy which represent primary shape information at the lower scales, but also subordinate energy which represent local shape information at higher differential scales. Thus, it provides a natural tool for multi resolution representation and can be used as a feature for shape classification. We addressed the three main processing steps in the MSED-based shape classification. They are namely, 1) image preprocessing and citrus shape extraction, 2) shape resample and shape feature normalization, 3) energy decomposition by wavelet and classification by BP neural network. Hereinto, shape resample is resample 256 boundary pixel from a curve which is approximated original boundary by using cubic spline in order to get uniform raw data. A probability function was defined and an effective method to select a start point was given through maximal expectation, which overcame the inconvenience of traditional methods in order to have a property of rotation invariants. The experiment result is relatively well normal citrus and serious abnormality, with a classification rate superior to 91.2%. The global correct classification rate is 89.77%, and our method is more effective than traditional method. The global result can meet the request of fruit grading.
A real time machine vision system for fruit size inspection was developed, which solved the problems such as fast processing the large amount of image information, improving system performance for real time dynamic image capture and processing capability, increasing precision of detection etc. For each fruit, four images were caught, and from which all the quality information of the whole surface were collected. Images were grabbed with a CCD camera (TMC-7DSP) and a frame grabber (Matrox Meteor II/MC), which is described in RGB space. The value of R/B was used as an index for image binary threshold after blurred image restoration. Median filter was used to denoise before edge detecting with Laplace Operator. A sphere fruit size-inspecting model was set up with a set of standard ball to calibrate the fruit size after the relative size of fruit, which was obtained with the method of partition edge point sets. The absolute error of the system was less than 1.1 mm and inspecting rate was over 31 fruits per second. That was this method can obtain fair inspecting speed, small absolute error, and filled the requirement of fruit automatic fruit sorting. But something is need to be paid attention, if shadow being in this vision system, it will arise big error when use partitions edge point, so it is needed to avoid the shadow.