Base On Computer Image Processing Technology For The Detection And Research Of Moldy Corn

This experiment took 4kinds of maize(Zhengdan958,Xianyu335,Yuan13,Jundan20) as raw materials, simulated the corn mildew in the specific environment(25℃, 60% humidity).The corn images were taken by parameters-fixed scan, under the help of computer image technology the color feature parameters of R,G,B,H,S,I were extracted. Then, we took some correlation analysises between these color feature parameters and the amount of corn fungi and the quantity of mycotoxins(Aflatoxin B1, Deoxynivaleno, Zearaleaone, Ochratoxin).Meanwhile, we established the discriminant model, distinguished the degree of moldy corn by BP neural network. The results showed that:(1)The amount of corn fungi showed the trend of increasing with the extension of time, the color of corn became more and more dark and the luster of corn became more and more bleak with the deepening of mildew. At 41 d, the samples deteriorated seriously,which were not able to go on. Depended on the amount of corn fungi, 4 kinds of corn samples were decided that 1-5d belonged to normal period, 7-11 d belonged to early mildew period, 13-33 d belonged to medium-term mildew, after 33 d belonged to late mildew.From the correlation analysises, there was a good correlation between the amount of corn fungi and the color feature parameters, the following models: the color feature parameter I value was introduced into Zhengdan958 model, the model was Y=1502.892-8.53 X I,R2=0.947; the color feature parameters R、I value were introduced into Xianyu335 model, the model was Y=5020.67-41.661 X R+20.199 X I,R2=0.934; the color feature parameters B、S、I value were introduced into Yuan13 model, the model was Y=﹣15602.569+463.54 X B+75209.492 X S-367.105 X I,R2=0.96; the color feature parameter I value was introduced into Jundan20 model, the model was Y=2696.205-15.445 X I,R2=0.823.(2)During the corn mildew, most of the quantity of mycotoxins showed a trend of increasing with the extension of time, among them, Ochratoxin in Zhengdan958、Zearaleaone, Ochratoxin in Xianyu335、Zearaleaone, Ochratoxin in Jundan20,they did not reach the detection limit of ELISA at the end of the experiment, so that these mycotoxins did not appear in the corresponding maize samples in this experiment.The variation trends of these color feature parameters R、G、B、I of maize images were consistent, all of them showed the downtrend with the experiment going-on, changes of H、S were not same, some showed increase(Aflatoxin B1 in Xianyu335,etc.),some showed decrease(Aflatoxin B1 in Jundan20,etc.),some almost were constant before and after experiment(Deoxynivaleno in Zhengdan958,etc.).The correlations between the quantity of mycotoxins and R、G、B、I value were extremely significant negative, the correlations between the quantity of mycotoxins and H、S value were different, the following models: the color feature parameter B value was introduced into Zhengdan958 Aflatoxin B1 model, the model was Y=441-3.088 X B,R2=0.834, the color feature parameter R、G value were introduced into Zhengdan958 Zearaleaone model, the model was Y=13660.795-234.464 X R +190.42 X G,R2=0.987, the color feature parameter R、G value were introduced into Zhengdan958 Deoxynivaleno model, the model was Y=742.015-10.234 X R +7.535 X G,R2=0.959; the color feature parameter G value was introduced into Xianyu335 Aflatoxin B1 model, the model was Y=462.893-2.738 X G,R2=0.842,the color feature parameter G value was introduced into Xianyu335 Aflatoxin B1 model, the model was Y=462.893-2.738 X G,R2=0.842, the color feature parameter R 、 H value were introduced into Xianyu335 Deoxynivaleno model, the model was Y=759.387-3.075 X R-3.702 X H,R2=0.958; the color feature parameter G value was introduced into Yuan13 Aflatoxin B1 model, the model was Y=389.275-2.28 X G,R2=0.962, the color feature parameter G、H value were introduced into Yuan13 Zearaleaone model, the model was Y=10030.308-72.704 X G+82.758 X H,R2=0.965, the color feature parameter I value was introduced into Yuan13 Deoxynivaleno model, the model was Y=766.277-4.508 X I,R2=0.814, the color feature parameter B、G value were introduced into Yuan13 Ochratoxin model, the model was Y=54.302-3.06 X B-0.84 X G,R2=0.946; the color feature parameter G value was introduced into Jundan20 Aflatoxin B1 model, the model was Y=409.876-2.173 X G,R2=0.896, the color feature parameter B 、 R value wereintroduced into Jundan20 Deoxynivaleno model, the model was Y=536.764-2.733 X B-0.736 X R,R2=0.98.(3)The BP neural networks were constructed by the color feature parameters for different degree of moldy corn, the results showed that: the BP neural network of Zhengdan958 was 6—15—4, the prediction rates of 21 different groups of mildew period corn samples(normal period 3, early mildew period 3, medium-term mildew 11,late mildew period 4) was 100%; the BP neural network of Xianyu335 was 6—15—4, the prediction rates of 21 different groups of mildew period corn samples(normal period 3, early mildew period 3, medium-term mildew 11,late mildew period 4) was 100%; the BP neural network of Yuan13 was 6—13—4, the prediction rates of 21 different groups of mildew period corn samples(normal period 3, early mildew period 3, medium-term mildew 11,late mildew period 4) was 95%; the BP neural network of Jundan20 was 6—14—4, the prediction rates of 21 different groups of mildew period corn samples(normal period 3, early mildew period 3, medium-term mildew 11,late mildew period 4) was 90%.