Kalman Peleg

Morphological control and mensuration of potato plantlets from tissue cultures for automated micropropagation

Automation in plant micropropagation can be greatly simplified if the propagated plantlets have some morphological properties that facilitate automatic chopping and subsequent inspection and classification of the pre-cut plantlet segments by machine vision as viable propagules. We were able to control the morphogenic pattern of in vitro-propagated potato plantlets by adding various concentrations of ancymidol to the nutrient solution. It was found that plantlets cultured in 0.25 mg l−1 ancymidol best fit the requirements for automated mass micropropagation; the mean internode length was sufficiently large (9–10 mm), the color contrast between leaves and stems was significantly enhanced, the stem was thicker than in the control treatment and the number of axillary buds per plantlet was maximized. Microtuber formation on segments isolated from plants cultured in 0.25 and 0.5 mg l−1 ancymidol media was enhanced shortly after transfer to tuber induction medium in vitro. On shoot segments from control plants, microtuber formation started after 24–28 days.Machine vision was used to evaluate the morphological and color changes in cultured potato plants. Geometrical and color features such as the number of buds, internode length and color contrast between leaf and stem were precisely measured and automatically logged. Features were measured that till now could only be observed qualitatively.

Classification of fruits by a Boltzmann perceptron neural network

Abstract Classification of fruits by machine vision is problematic in two respects: (a) Most of the sorting criteria are “fuzzy”, because the class membership can not be quantified precisely. The reference classification is subjectively determined by a trained panel of inspectors, that often disagree as to the class of the fruit. (b) The statistics of the classification criteria vary with harvest time and from orchard to orchard, so the classifier must be easy to re-train. Using digital color imaging hardware and a BPN based classifier we developed a system for sorting fruits that can address these problems. It naturally accepts “fuzzy” or “soft” labeling at train-time and can be tuned to provide various levels “soft” or “fuzzy” decisions at run-time. The power of the BPN was demonstrated by a synthetic dataset, indicating that the BPN can create intricate non-linear discriminant functions, even when the classes are noncontinguous and the training set is relatively small. Simulated sorting experiments of apples, into “red” and “green” categories showed that the system can emulate the decisions of a panel of human sorters quite well.

Machine identification of buds in images of plant shoots

Tissue cultures find increasingly widespread applications for cloning of many plants. Commercial propagation by tissue cultures is limited to ornamental plants, because the cost of skilled labor required cannot compete with conventional propagation methods. To cut down the cost, some automation is essential. A cost-effective approach is to chop the plantlets into segments on a conveying production line while using machine vision for identifying and locating the number and positions of propagation organs in images of the plantlet segments. Plantlet segments without propagation organs will be rejected, while segments with viable buds will be selected for subculturing. To this end, a machine-vision-controlled automatic subculturing system for potato tissue cultures is proposed as a simpler and more cost-effective solution than the popular trend of imitating the manual sub-culturing task by a robot. A simple and relatively fast image-processing algorithm particularly suitable for classification of potato tissue cultures, was developed. In lieu of the general Medial Axis Transform approach, this specialized algorithm takes advantage of the inherent difference between the geometrical shape and gray scale levels of the stems and the leaves as well as of the rather simple connectivity rules of attachment between them. The results indicate that machine inspection and classification of tissue culture plantlets is possible, but considerably more work needs to be done before this technique is fully developed for automating tissue culture processes.

Vibration modes of spheroidal fruits

The motivation for these studies is that the frequency response characteristics of fruits and vegetables is strongly correlated to their firmness and maturity. Adequate understanding of the various vibration modes is essential for practical application of this phenomenon to efficient sorting of fruits by firmness and maturity. The geometrical deformations associated with different vibration modes of fruits, as described in previous studies, were assumed to follow certain theoretical models. However, no conclusive proof was given, based on measuring these deformations experimentally. In this paper we describe an experimental procedure and provide a set of rules for studying and recording the modal shapes of vibrating spheroidal fruits. Experimental work was carried out using relatively soft fruits (oranges) and firmer fruits (apples). A lumped parameter model based on a multidegree-of-freedom system (MDFS) is proposed, in which the spheroidal fruit is divided into three vertically vibrating masses, interconnected by springs and dashpots. Model parameters are calculated from experimental frequency response curves. The model imitates the first three resonance modes and provides essential mechanical parameters of vibrating fruits.

Firmness Indexes of Viscoelastic Bodies by Vibration Testing

In many cases it may be beneficial to measure the overall textural properties of an object by a single number rather than focusing on particular parameters in a rheological model. Intuitively, the concept of “firmness” may serve this purpose. Firmness may be used to classify fruits and vegetables by maturity, foods by textural properties, foamed plastics by their cushioning properties, etc. The definition of firmness has hitherto not been rigorously given. Using a previously developed general rheological model, which is suitable for quantifying the behavior of solids made of linear and nonlinear viscoelastic materials, theoretical and experimental groundwork is laid, whereby an object may be subjected nondestructively to a series of minute deformations by vibration excitation, in conjunction with computerized determination of various indexes for automatic classification by firmness. Development of several alternative firmness indexes is described and their relative utility for automatically classifying vi...

Container dimensions for optimal utilization of storage and transportation space

Abstract A computer program is described whereby a printout is obtained of all feasible container dimension and stacking pattern forms, for say more than 90% volume utilization of a given master container. The program can be used as a design tool of container (package) dimensions for efficient utilization of master containers, pallets, storage rooms, maritime and aviation containers.

ADAPTIVE CLASSIFICATION BY NEURAL NET BASED PROTOTYPE POPULATIONS

We have developed an algorithm for unsupervised adaptive classification based on a finite number of “prototype populations” with distinctly different feature distributions, each representing a typically different source population of the inspected products. Intermittently updated feature distributions, of samples collected from the currently classified products, are compared to the distributions of pre-stored prototype populations, and accordingly the system switches to the most appropriate classifier. The goal of our approach is similar to the objectives of the previously proposed “Decision Directed” adaptive classification algorithms but our solution is particularly suitable for automatic inspection and classification on a production line, when the inspected items may come from a finite number of distinctly different sources. The recognition of prototype populations as well as the classification task proper may be implemented by conventional classifiers, however neural networks (NN) are advantageous in two respects: There is no need to develop separate mathematical models for each classifier because the NN does it automatically during the training stage. The parallel structure of NN has the potential for very fast classification in real time, if implemented by dedicated parallel hardware. This is particularly important for high speed automatic sorting on a production line. The practical feasibility of the approach was demonstrated by two applied examples, wherein two prototype populations of apples are recognized and sorted by size and color derived by machine vision. Three “Boltzmann-Perceptron Networks” (BPN) were used, one to recognize the prototype populations, while switching between the other two, for optimally classifying apples into two size and color categories. It is shown that misclassifications by adaptive classification are reduced, in comparison to non-adaptive classification.

Optimal adaptive classification of agricultural produce

Abstract Automated grading of agricultural products suffers from the time varying, error-prone feature space used. Classification errors can be minimized by choosing appropriate class boundaries. However, changes in the statistics of the features in the produce stream cause these optimal boundaries to change. Modern sorting machines use a multitude of features which complicates the optimization of the class boundaries. Automated optimization can be accomplished through identification of distinct, discrete populations within the produce stream, and training of an optimal classifier for each population The adaptive algorithm has been simulated in software and aspects of the algorithm are explored.

Optimal scale translations in noisy measurement systems

In engineering, medicine, biology and agriculture, it is often desired to replace an invasive or slow measurement method, by nondestructive, faster or less expensive methods. The inevitable question is whether the two measurement methods are interchangeable. To answer this question, the common practice is to use linear regression based equations, as scale translation rules. It is shown that this approach is not optimal, when both measurement methods are noisy. Accordingly, a new approach for method comparisons is proposed, by high fidelity translation of the readings taken on the scale of one test, to the scale of another test, and vice versa. The proposed scale translation mode is based on minimizing the sum of squares of the differences between the absolute values of the fast Fourier transform (FFT) series, derived from the readings of the compared measurement methods. Whereas regression methods attempt to find the parameters of a line that provides the best fit to the observed data pairs, the FFT equalization method strives to find the parameters of a line that can render the difference between the translated readings as close to zero as possible. The line taken is illustrated by a comparative study on several artificial datasets of linearly related paired X, Y readings, with various levels of measurement noise. Quality criteria were developed for quantitative comparison of linear regression based, scale translation models versus the new method, while using the results from the artificially generated datasets for illustration. The comparisons indicate that scale translation by the FFT equalization method is optimal in terms of these quality indexes.

Classification of apples with a neural network based classifier

Abstract 352 apples were classified as green or red by three human experts. Each apple got a probabilistic (fuzzy) membership in the red and green sets according to the number of votes. The majority decided the actual classification. This fuzzy data, and features such as color hue, saturation, and intensity, and the ratio between red and green areas, for the first 30 (or 60) apples were used to train a Boltzmann Perceptron Network (BPN). The remaining fruits were then classified by the BPN. It was found that hue is the dominant feature. The misclassification was less than 15%, similar to a Bayesian classifier trained on the same data. The advantages of the BPN seem to be that it allows fuzzy input data, and many features.