Research And Application On Apple Image Feature Extraction And Classification Algorithms

The study and development of apple's automated real-time grading system to select high quality fruits, created for the country to take foreign currency, in our country has an important economic value and broad application prospects. The current studies mostly are apple's quality inspection of agricultural products at home and abroad. In the apple quality inspection, abroad some quality of detection projects(such as size, shape, color, surface defects, etc.) have been commercialized, and can achieve real-time speed. In china, the quality of detection of apple from the 90's began, and it stayed on external quality inspection, and much did not reach level of real-time detection classification. China's apple production in the whole agricultural production accounts for a large proportion, and is an important export product. However, post-harvest handling is not enough to make exports difficult to ensure the quality of apples. So it lacks competition in the international market. The reasons first and foremost are that a means of detection and sorting gets behind, and apple's classification in china was still largely manually completed. Disadvantages of artificial classification are: a large quantity of labor, low productivity, difficult to achieve grading standards, instability of classification accuracy. Because it difficult to accurately distinguish by the human visual in apple grading standards coloring area and defect area measurements, and the prolonged use eye will cause fatigue and emotional instability, resulting in classification error fluctuations.In this paper we use computer vision as the main means. We use pattern recognition technology to detect apple scar, and we focus on feature extraction of apple scars and classification recognition technology. The main contribution of this paper is following area.①In the apple's image pre-processing stage, this paper adopted a median filter to remove noise to improve image quality, and it reduces the computational complexity to improve the implementation of the follow-up algorithm for the speed and convergence speed.②In the feature extraction stage, this paper discussed the extraction method of apple scar local feature, including color features, texture features, gray level co-occurrence matrix as well as the scale invariance feature(SIFT) transforms extraction methods, and in this paper we highlighted used in the local features based on scale invariant(SIFT) transformation method. ③In the classification stage, in this paper we studied application of support vector machine(SVM) in apple scar classification, in order to identify the accuracy, we identify two steps, firstly the image should be divided into a small window and detect windows, then we combined the small window, further filling some empty that may exist by morphological processing, and then detecting key points of apple scar in large scale space, and finally we marked key points of the region,through this method we can very well recognize apple scar.Finally, we summarized the full contents, and several questions for further research and exploration are proposed.