Study On Automatic Grading Methods Of Fresh Corn Ear

Fresh corn ear contains many kinds of amino acid, vitamin, trace element and fatty acid etc, and possesses certain medical and health care function. With more attention paying on the nutritive value, demand of fresh corn ear increases gradually. Corn raw material sampling and online quality detection were accomplished by human. A large amount of manpower and material resources were expended using artificial detection. Because of heavy intensity of labour, detection results are greatly influenced by fatigue and other environmental factors. Corn production is faced with many opportunities and challenges for expanding domestic demand and enhancing innovation of the 12^th five-year plan. Based on problems in corn ear grading, sampling detection and online quality grading methods developed by computer vision, sensor and neural networks technology in this dissertation are as follows :1.Quality analysis detection hardware system was designed including trigger device, acquisition device, transport mechanism, actuator and pressure detection device. In order to acquire over all image data, fresh corn ear was forced to go forward and rotate around its own centerline. Artificial press simulation was realized using the pressure detection device including pressure sensor and actuator, and the pressure data correlated to corn maturity could be achieved. Rapid automatic detection hardware system was designed including trigger and acquisition devices, transport mechanism and actuator to realize real-time online detection of fresh corn ear.2.Grading of fresh corn ear appearance quality based on ear shape and defection was achieved using computer vision firstly.Ear image was captured by quality analysis detection hardware system. In HSI color model, bare tip was identified preliminarily through an automatic threshold method and eight neighborhood templates, then its starting and ending positions were determined and removed using Laplace transformation and projection method. Defects of fresh corn ear were identified by the first order differential operation on H and the defect ratio Q was calculated. Ear shape parameters were extracted as follows: ear length L, maximum diameter W, aspect ratio r and rectangle factor R. The average errors of bare tip position, ear length and maximum diameter were 2.27, 1.96 and 0.54mm, respectively.The mistake rate of defect ratio was 3.00%. General regression neural network was developed for grading. Five characteristic parameters Q, L, W, r and R were used as inputs, and the grading average ratio was up to 95.56%. Based on the rapid automatic detection hardware system, ear length and maximum diameter were obtained by scanning method: grading speed was more than 60 ears per minute, and grading accuracy was more than 90%.3.The maximum pressure data was extracted by pressure detection device. The inertia moment of texture information was extracted from part ear image which was captured by quality analysis detection system. Combined with maximum pressure data and inertia moment, system cluster analysis was carried out through between-class connection method and squared euclidean distance. According to the results, maturity was determined to be 3 grades which is consistent with the production requirements. In order to improve grading accuracy, eleven color characteristics that are were extracted from whole ear image. Eleven color characteristics were optimized and screened by principal components analysis method. The first and second principal components were applied to represent the eleven color characteristics in the grading. Probabilistic neural network was established, and inertia moment, maximum pressure, the first and second principal component values of color characteristics were used as the inputs of the network for maturity grading of fresh corn ear. The grading accuracy was 96.67%. Based on rapid automatic detection system, fresh corn ear online grading was implemented according to (R|-),(B|-) and b in RGB color model. In HSI color model, online grading was implemented according to E value. Grading speed of the two methods was more than 60 ears per minute, and the grading accuracy was more than 90%.4.Partial image of fresh corn ear was captured by quality analysis detection system. In HSI color model, wavelet decomposition of H component was carried out. According to (S|-) j value, decomposition level of wavelet was determined to be 2. Wavelet coefficient was extracted from sub-band of LL₂, and its mean value A, variance V, mean value of energy E_a, and color entropy E_t were used as color characteristics. In RGB color model, eleven texture characteristics were extracted which were grads mean square error(W_Gds), grads mean(W_Gdm) and small grads dominance(W_Sga) of HH1 sub-band,grads entropy (W_Gde) of HL₂ sub-band, gray entropy(W_Gye),big grads dominance(W_Lga),grads mean(W_Gdm),entropy(W_He),gray mean square error(W_Gys) and small grads dominance(W_Sga) of LH₂ sub-band. Dimension reduction was implemented for the 14 characteristics in using principal components analysis method. The first four principal components were applied to represent the fourteen color and texture characteristics. Probabilistic neural network was established and the four principal component values were used as the inputs for maturity grading. The grading accuracy was 93.33%. p₁ value of energy spectrum was extracted from partial image through Fourier transform. Combined with the four principal component values and p₁ , probabilistic neural network was used to implement maturity grading, and the grading accuracy was 95.56%. Compared with the wavelet decomposition method, the grading accuracy of compound method was higher, but processing time was longer.5.Based on VC++6.0 software development platform and SQL Server data base programming technology, fresh corn ear quality grading software system was developed. three systems were developed which were quality analysis detection software system, rapid automatic detection software system and data base system. Incorporating with hardware devices, sampling and online detection was accomplished. Data of fresh corn ear could be saved, inquired and analyzed.In conclusion, key techniques of fresh corn ear appearance quality and maturity detection investigated in the dissertation provids theoretical foundation and technical support for quality detection before further processing. Through productional application, sampling and online detection using quality analysis detection system and rapid automatic detection system were established in much shorter time with higher accuracy. Rapid automatic detection system is suitful to real time and on line detection grading. Quality analysis detection system is for multi- targets sample detection of corn ear.