Study Of The Information Acquisition And Processing Technology Based On Embedded Computer Vision

The existing machine vision systems are inconvenient to carry to the field for work. The expensive price and complex operation restrict it from popularization to the farmers. The embedded machine vision technology is a kind of machine vision technology using the embedded computer as data processing, and its software and hardware can be cut according to its practical application. It is the expansion and extension of the machine vision technology. Study of the information acquisition and processing technology based on embedded system computer vision has practical application in agriculture, forestry, animal husbandry and fishery which are far away from cities with limited Internet service and power supply. Thus, related research of machine vision technology in handheld devices has practical sense.In this context, a handheld machine vision system was developed using the machine vision technology. The main researches were focused on the development of handheld machine vision system and the test of its adaptability in crop growth information collection and data process. Fruits, plant leaves and citrus were used as experimental subjects to evaluate the adaptability of the algorithms to calculate the size of fruits, the area of leaves and the yield of citrus, and good results were obtained in this research. The main conclusions are as follows:(1) The developed handheld device vision system was adopted to study the processing algorithm of the handheld device to achieve higher speed. The procedures include dimension reduction for array, usage of integer arithmetic to simulate floating point, combination with computer principle, algorithm transmission from multiplication and division to shift operator, rational definition of data type in program, algorithm transmission from multiplication and division to table lookup, application of self-developed algorithm instead of that provided by development environment. According to the results of image grizzled processing, using this methods as mentioned above makes a significant improvement on the speed of software with more than20times. The methods were high flexibility and practical value.(2) Two handheld machine vision systems were developed. This system used mobile phone development module to replace the computer in computer vision system to set up a handheld machine vision system for data process. The inclination measurement module was used to measure the angle between the camera and the target. The distance measurement module was used to measure the distance between the camera and the target. The other one was Android based smart cellphone19300. The application softwares integrated with machine vision were written in Java. The app’s algorithm applicability verified that the portable vision technology possesses stability and repeatability. All above extends application of computer vision in agriculture.(3) The developed handheld device vision system was performed to discriminant the quantity of cluster fruits. The paper introduces an improved Freeman chain code to discriminant the number of fruits. It adds three new elements namely "S’、"8"、"9", while the "S" is the starting point of clustering area edge,"8" is the turning point, and "9" is the lowest point of image contour. After the coding of edge image, the number of "8" is used for the discrimination of fruit number in clustering area. When the method is tested on Gannan orange and Yantai apple,100%discriminant rate is accomplished in two, three and four clustering case, in five clustering case can reaches above60%, but it is very difficult to correctly discriminant six clustering case. This method can extend to other clustering cases in fruit that has regular contour, like pear, tomato etc. and has theoretical and practical meaning in enhancing pre-harvesting fruit yield estimation.(4) The developed handheld device vision system was applied to estimate the citrus yield of a single tree two weeks before harvest. A semi-automatic self-adjusted segmentation algorithm was promoted for on-site application. It utilizes morphology filter for edge smoothing in clustering area, then improved Freeman8neighborhoods for clustering area fruit number counting calculates the total number of citrus. This algorithm has been used for Ganan orange yield estimation and reaches90%accurate rate. This method facilitates orchard pre-harvesting yield distribution mapping, and provides technical solution for further yield estimation.(5) The developed handheld device vision system was used to research on the vision algorithm for information collection and processing methods of crop growth. In this research, we put forward a semi-automatic adaptive segmentation algorithm for handheld device. During the application of leaf area measurement, after artificial selection of leaves to be measured, the boundary of the leaf image displayed on the touch screen is drawn as a closed circle. And then, Canny edge detection algorithm is used to obtain the information of the edge pixel position of the leaf followed by median filtering and binarization (using Otsu threshold algorithm). After the image reconstruction, the edge pixel would be inserted into the image, and the pixels within the edges would be filled. Results show that the display error is within1%which means this method ensures the integrity of the research object edge while eliminating the white noise in the image to improve the accuracy of the measurement results.