Research On The Technology Of Hami Melon Maturity Detection System Based On DSP

Hami melon is well known as’ The king of the melon’,which enjoying a good reputation at home and abroad for advantages of high sugar content and a sweet smell greets the nose. Hami melon not only has a peculiar taste but also good for nutrition and has high officinal value. At present,Hami melon’s uneven quality lead to high quality but low price in the world market due to its low post-harvest treatment and backward processing means. Thus, one perfect quality detection system is urgently needed.In this paper, interdisciplinary knowledge is used such as digital image processing, electronic information science, and pattern recognition, choose’Gold queen’Hami melon in Xinjiang for the research object. This thesis researches Hami melon maturity real-time detection technology which sets off from image color and texture feature extraction angle. This research predicts the Hami melon maturity based on color and texture features creatively and builds a model of support vector machine to predict maturity, then optimize the model and transplant it to Digital Signal Processor to achieve the goal of predicting Hami melon’s maturity in real time and fleetly. The main results as follows:1. Hami melon images acquisition system and DSP video processing platform was builded. The former were used to collect the image of Hami melon, the later were used to real-time detection.Different denoise and background segmentation method are compared in this research. Results of compare analysis indicated that median filtering and arithmetic channel operations segmentation is a better fit for Hami melon mentioned in this paper.2. Hami melon color feature of two different maturity from two color space was extracted and analyed.The results from range of color features showed that:1) compared the color features of two maturity, the mean H value of 90 percent of ripeness is 0.120,which is less than H value of 80 percent of ripeness(0.128);the mean G value of 90 percent of ripeness is 188.740,which is less than H value of 80 percent of ripeness(197.830); the RG difference value of 90 percent of ripeness is 32.212,which is less than RG difference value of 80 percent of ripeness(39.862); the RG sum value of 90 percent of ripeness is417.340,which is less than RG sum value of 80 percent of ripeness(39.862); the R-G/R+G+B value of 90 percent of ripeness is 0.080,which is more than R-G/R+G+B value of 80 percent of ripeness(0.063). Ripe Hami melon has less mean H value, mean G value, RG difference value, RG sum value and more R-G/R+G+B value.2) mean B value, mean S value, mean I value have no significant relationship with maturity.3. Hami melon texture feature of two different maturity was extracted and analyzed. The results from range of color features showed that: 1) contrast, smoothness and dependency have close correlation to maturity. Compared the texture features of two maturity, the contrast of 80 percent of ripeness is in the range of 766.31 and 1207.27, while the contrast of 90 percent of ripeness is in the range of 974.52 and1675.33, contrast increase with maturity.The smoothness of 80 percent of ripeness is in the range of 3.70 and 3.80, while the contrast of 90 percent of ripeness is in the range of 3.58 and 3.76, Smoothness increase with maturity. The dependency of 80 percent of ripeness is in the range of 2909313 and 4476398, while thedependency of 90 percent of ripeness is in the range of 3131968 and 4047242, smoothness decrease with maturity.2) energy and entropy have no significant relationship with maturity.3. Support Vector Machine model was builded based on the relationship between color feature and maturity. Final model accuracy was 97.22% after parameters setting and optimizing.A video real-time detection platform based on Digital Signal Processor was builded, include periphery module such as IIC bus, storage space map, video capture, video buffer, video display and so on, Hami melon maturity detection arithmetic was designed. Final classification accuracy was 94.34% after arithmetic optimizing and program target device due to Hami melon maturity detection algorithm.