No abstract

Abstract:Apples should be quickly and correctly detected from their surroundings for the apple harvesting robot. The basic color feature was extracted from FuJi apple tree images and analyzed by the statistical analysis method. Accordingly, a new apple detection method was proposed to position the centroid of picking apples. The color difference was used to segment apples from their surroundings. Then the picking apples were chosen by area parameter. After that, the conglutinated apples were segmented by bidirectional scanning line algorithm. Finally, all of picking apples were positioned by their circumdiameter matching algorithm. The experimental result showed that the correct classification rate of apple fruit achieved 90%.

Abstract-The apple harvesting robot should have two eyes to sensor the location of picking apples. The binocular machine vision system used two Canon digital cameras was built, instead of digital video used as usual. Thus, the digital camera vision system had higher resolution and superior performance than a digital video vision system, which could capture a Jpeg image with 3456*2592 pixels, its field of view included the whole apple tree. For the Fuji apple tree, it is an obvious color difference with ripe apples and their surroundings of leaves and braches. Therefore, the Drg-Drb color index was used to segment apples from their surroundings. Then the mistaken classified background regions was deleted by the area filter, and he picking apples were similarly chosen by area parameter. After that, the conglutinated apples were segmented by the bidirectional scanning line algorithm, which were scanned from the horizontal and vertical direction. Finally, all of picking apples were positioned by their circum-diameter matching algorithm. The experimental result showed that the correct classification rate of picking apple fruit achieved 90%.

Weed detection is a key problem of spot spraying that could reduce the herbicide usage. Spectral information of plants is very useful to detect weeds spectrograph-based weed detection system is too high. Therefore, the main objective of this study was to explore a method to classify crop and weed plants using the spectral information in the visible light captured by a CCD camera. One approach to weed classification was to directly use of G and R component of RGB color space. Another was to utilize the spectral information among the green band that hue was regarded as wavelength, and saturation was represented as reflectance. The result of statistic analysis showed that both of them using the G-R and H-S optimized segmentation line of crop and weeds could be used to detect weed (lixweed tansymnustard) from wheat fields. Moreover, the method of using the H-S optimized model could avoid the affect of lighting.

No abstract