| The color of farm product and food is the very important quality target of product. While the vegetable color is more important, because the vivid vegetable color can not only stimulate the consumers' desire of buying, but also accumulate the people's food desire. It's more important that color is an important target that whether the vegetable is fresh.The color of objects is a psychology physical quantity. It has relation with the object itself and human color vision. The appraising and comparing of food color can only depend on human's (he sometimes is the professionals who have some kinds of authority) subjective observation to draw a conclusion. But this method can be affected much by human's subjective factors, so it can hardly draw an objective conclusion.Computer vision can not be affected by human's physiological and psychological factors, so it can precisely quantize the color of farm product, so it can get comparatively objective appraise and the result of appraising is stable, standard and objective. So it is a new way of objective appraising of color sense organ.Computer vision can be simply understood that cameras replace the human's eyes and computer replaces the human's brain. So the recognitions and explanations to ambient environment can be accomplished. Its hardware components are the computer image processing system which includes computer, cameras, image processing card (which converts the simulant image signals into digital signals), lamp-house and etc. Its fundamental principal is that it can acquire the two-dimension image information by using cameras and can convert the image information into digital image that the computer can accept by using signal conversion.Due to the nonlinear of the conversion from the RGB color space to the XYZ color space, we need establish a nonlinear conversion relation. Neural networks have fairly well capability of nonlinear map, especially that BP neural networks have 3 characters:The BP neural networks that have two latent layers, can accomplish arbitrary map.It has simple structure and can be easily accomplished.It has rapid computing speed.The principle of BP arithmetic: It has two processes that forward computing (forward propagation) of data stream and backward propagation of error signals. During forward propagation, the direction of propagation is input layer→latent layer→output layer and the neural of every layer can only affect the neural which belong to the next layer. If it cannot obtain the desirable output, it turns to the process of error signals backward propagation. During the two processes alternately performing, it performs strategy of error function grads declining, and it search a group of value vector by performing dynamic iteration. So the network error function get to minimum and the process of information picking-up and the memory is complete.In this paper we set up a suit of system that can detect food color and focuses on having research on several key technologies as follows:We have analyzed the RGB, XYZ, L*a*b*, HIS color space. And we have indicated the defaults of RGB color space and XYZ color space. L*a*b* color space is better than XYZ color space for comparing and detecting color difference, so we use L*a*b* color space to compare and detect color difference. At the same time, we have illustrated that it is nonlinear that conversion relation between the RGB color space and XYZ color space, so we need a nonlinear conversion relation by using neural networks. The image captured by the cameras is expressed by the RGB color space. Because we can not converse the RGB color space into XYZ color space directly, so we should converse the RGB color space into XYZ color space first, then we converse the XYZ color space into L*a*b* color space.We have set up a computer version system that includes computer, color CCD camera, image capturing card and illumination system. The work of gathering image color is accomplished by software that is programmed by using VC++6.0. The software can capture the RGB col...
|
PDF Full Text Request