CN106442525A - Online detecting method for wizened defects of insides of walnuts - Google Patents

Abstract

The invention discloses an on-line detection method for internal shriveled defects of walnuts. An industrial CCD camera is used to obtain images of walnuts under dynamic conditions, and binary processing is performed to obtain the projected area. The weight information of the walnut under the dynamic condition is obtained by combining the acceleration sensor; then the walnut weight prediction model obtained by the regression analysis of the projected area and weight of the walnut is used to calculate the predicted weight of the walnut, and calculate its relationship with the real quality of the walnut The relative error between them, using the relative error as the discriminant threshold to judge whether the detected walnut is shriveled walnut, and crack the shell to verify the discriminative accuracy; then perform regression analysis on different discriminant thresholds and their corresponding discriminative accuracy; finally use the golden section Search the fitting function between the discrimination threshold and the discrimination accuracy to find the best discrimination threshold. If the relative error of the detected walnut weight is greater than this threshold, it will be judged as shriveled walnut. If the relative error of the detected walnut weight is less than or equal to This threshold is used to identify normal walnuts; this method can realize the non-destructive detection of shriveled walnuts only by using the image projection area of walnuts and walnut weight information. The speed and accuracy are suitable for large-scale production in factories, and can be used for online detection of internal shriveled defects of various nuts such as walnuts.

Description

On-line detection method for walnut internal shriveling defects

technical field

The invention relates to a method for detecting internal quality of nuts, in particular to an on-line detection method for internal shriveled defects of walnuts.

Background technique

Walnut shriveling defect is a common quality problem of intact walnuts in the market, which seriously affects the quality and sales price of commercial walnuts. The main causes of walnut shriveling defects under natural conditions are: the limitation of walnut growth environment conditions, such as low soil nutrients and water shortage; diseases and insect pests during the growth of walnuts, etc. Therefore, the walnut shriveling defect inevitably exists in the whole walnut products sold in the market. At the same time, walnut shriveled defect is different from other defects, it belongs to internal quality problem, shriveled walnut and normal walnut are indistinguishable in appearance. Under natural observation conditions, the appearance of the two is basically the same, but after breaking the shell, it can be found that there are serious quality problems inside the shriveled walnut. Based on the above reasons, the present invention proposes an on-line detection method for shriveled walnuts based on fusion of force sensing and visual information. Only digital image information and weight information of walnuts can be used to realize fast online non-destructive detection of shriveled walnuts.

At present, there are few studies on the detection of internal defects of walnuts, and most of them are researches on the detection of defects under static conditions. There are no relevant reports on the online detection of internal defects of walnuts in China. Huang Xingyi et al. used soft X-rays to obtain images of walnuts to distinguish empty shells, damaged walnuts, and normal walnuts. The accuracy of the test was high, but this method had radiation risks for operators, and the food safety risk was unknown; Jensen et al. used near-infrared spectroscopy to identify The acetaldehyde content of walnut kernels was analyzed by partial least squares regression, r2=0.72; Li Bin et al. initially explored the feasibility of using terahertz spectroscopy to detect pecan pests, and found that due to the higher water content of living pests and the higher water content of terahertz Spectrum has strong absorption characteristics of polar molecules such as water. Through comparison with pecan slices, it is found that live insect pests have very obvious spectral absorption characteristics, which are different from normal walnuts. Although near-infrared spectroscopy and terahertz spectroscopy are sensitive to moisture and can detect moisture inside walnuts and living pests, etc., there is little water in deteriorating thin-skinned walnuts such as shriveled and empty shells, so near-infrared spectroscopy and terahertz spectroscopy cannot detect Deteriorated thin-skinned walnuts such as shriveled and empty shells are effectively detected. Moreover, the cost of the above technical equipment is high, the amount of information and data is large, and the real-time performance is poor, most of which are difficult to meet the requirements of online production.

Contents of the invention

The purpose of the present invention is to provide a low-cost, high-accuracy, fast and non-destructive on-line detection method for walnut internal shriveled defects.

The purpose of the present invention is achieved through the following technical solutions:

The steps of the method are as follows: obtain the image and weight information of the walnut under dynamic conditions, perform binary segmentation and morphological processing on the image to remove the background to obtain the projected area of the walnut, and perform regression analysis on the projected area and weight of the walnut to obtain the optimal Walnut quality prediction model, through this model to predict the quality of walnuts, and calculate the relative error between it and the real quality of walnuts, and then use the golden section optimization algorithm to find the best discrimination threshold, which is used as the discrimination of whether the walnuts are shriveled walnuts Threshold to judge whether the tested sample is shriveled walnut.

1) the walnut dynamic image: take the black synchronous conveyor belt as the background, utilize the camera external trigger collection mode and the weight information synchronous collection of walnut;

2) The collection of the walnut weight information: digitally filter the walnut weight signal collected under the dynamic condition of the cantilever beam load cell by using a digital acceleration sensor, so as to obtain more accurate walnut weight information.

3) The walnut image binarization segmentation and morphological processing to remove the background: extract the R component of the walnut RGB image, perform background segmentation with 50/255 as the segmentation threshold, and then use the median filter and continuous open operation to remove noise to obtain Go to the walnut image for the background.

4) described walnut projected area: collect the image of coin under the same condition, obtain its background image and calculate the standard area of a single pixel through the same processing method, and count the pixels occupied by walnut in the walnut image of background number, calculate the projected area of the walnut.

5) The relative error of the walnut quality: with the obtained walnut projected area as the independent variable, the walnut weight measured under the dynamic condition is the dependent variable to carry out regression analysis and obtain a linear correlation between the two, and use this as the walnut weight prediction The model calculates the predicted value, and the relative error of the walnut weight is obtained by dividing the absolute value of the difference between the predicted value of the walnut weight and the test value by the test value.

6) The golden section optimization algorithm seeks the optimal threshold: take the relative error of the weight of the tested sample as the independent variable, and the accuracy rate of discrimination when each relative error is the threshold is the dependent variable, and regression analysis is carried out to obtain the difference between the two. There is a Pulse function relationship. This function is a unimodal function. The golden section method is used to calculate the maximum value in the interval of the independent variable. At this time, the value of the corresponding independent variable is the optimal threshold.

7) The shriveled walnut discriminant method: the walnut image under the online condition is processed through digital image processing and calculated to obtain the walnut projected area, predicts the walnut weight with the projected area, and obtains the relative error of the walnut weight with the walnut test weight calculation, and obtains the optimal Compared with the threshold value, if the relative error is greater than the threshold value, it is judged as a shriveled walnut, otherwise it is a normal walnut.

Compared with the existing technology, the present invention realizes the on-line detection of walnut shriveled defects based on the material characteristics of the walnut itself and based on the method of fusion of force sensing and visual information, which is low in cost, fast and effective, and is suitable for industrial production. The invention is accurate, effective, economical and practical, and has strong adaptability, and is a method capable of well realizing on-line detection of walnut shriveled defects.

The beneficial effects that the present invention has are:

The invention can realize online non-destructive detection of shriveled walnuts only by using digital image information of walnuts and walnut weight information. The method is simple. Compared with the existing technology, the detection cost is greatly reduced, and the detection speed and accuracy are improved. It is suitable for Factory scale production. The invention can be used for on-line detection of internal shriveled defects of various nuts such as walnuts.

Description of drawings

Fig. 1 is a flow chart of the implementation of the on-line detection method for internal shriveled defects in walnuts according to the present invention.

Fig. 2 is a schematic structural diagram of the device of the present invention.

Fig. 3 is the walnut image obtained after image processing and the number of pixels it occupies after removing the background.

Fig. 4 is the fitting relationship curve between walnut projected area and walnut weight.

Fig. 5 is the fitting curve relationship between the discrimination accuracy rate and the discrimination threshold under each threshold.

In the diagram: 1 is an industrial CCD camera; 2 is a synchronous conveyor belt; 3 is a DC motor; 4 is a digital acceleration sensor; 5 is a spacer; 6 is a base; 7 is a weighing sensor; 8 is a bracket; 9 is a walnut sample; 10 is the light source.

detailed description

Embodiment: FIG. 1 shows a flow chart of an online detection method for internal shriveled defects in walnuts. First of all, the walnut sample is placed on the synchronous conveyor belt 2, and the clear image is obtained by the industrial CCD camera 1, and it is binarized to obtain its projected area, and the load cell 5 and the digital acceleration sensor 6 are used simultaneously The weight information of the walnut under the dynamic conditions is obtained in a combined way; then the walnut weight prediction model obtained by the regression analysis of the projected area and weight of the walnut is used to calculate the predicted weight of the walnut, and calculate the relationship between it and the real quality of the walnut. Relative error, using the relative error as the discrimination threshold to judge whether the detected walnut is shriveled walnut, and crack the shell to verify the discrimination accuracy; then perform regression analysis on different discrimination thresholds and their corresponding discrimination accuracy; finally use the golden section method to search for discrimination The fitting function between the threshold and the discriminant accuracy finds the best discriminant threshold. If the relative error of the detected walnut weight is greater than this threshold, it is judged as shriveled walnut. If the relative error of the detected walnut weight is less than or equal to this threshold, then It was identified as normal walnut.

Claims (2)

1. An online detection method for walnut internal shriveled defects, characterized in that: the method steps are as follows: obtain image and weight information under the dynamic condition of walnut, carry out binary segmentation and morphological processing to the image and remove the background to obtain After the projected area of the walnut, perform regression analysis on the projected area and weight of the walnut to obtain the optimal walnut quality prediction model, predict the quality of the walnut through this model, and calculate the relative error between it and the real quality of the walnut, and then use the golden section to optimize The algorithm finds out the best discrimination threshold, which is used as the discrimination threshold for judging whether the walnut is shriveled walnut, and judges whether the tested sample is shriveled walnut. 2. a kind of on-line detection method that is used for walnut internal shriveled defect according to claim 1, is characterized in that: 1) The image under the walnut dynamic condition: take the black synchronous conveyor belt as the background, utilize the external trigger acquisition mode of the camera and the weight information synchronous acquisition of the walnut; 2) The collection of walnut weight information under dynamic conditions: digitally filter the walnut weight signals collected by the cantilever beam load cell under dynamic conditions by using a digital acceleration sensor, so as to obtain more accurate walnut weight information. 3) the binarized segmentation of the walnut image and the morphological processing to remove the background: extract the R component of the walnut RGB image, carry out background segmentation with 50/255 as the segmentation threshold, and then utilize median filtering and continuous open operation to remove noise, Get the walnut image with the background removed. 4) described walnut projected area: collect the image of coin under the same condition, obtain its background image and calculate the standard area of a single pixel through the same processing method, and count the pixels occupied by walnut in the walnut image of background number, calculate the projected area of the walnut. 5) The relative error of the walnut quality: take the gained walnut projection area as an independent variable, and carry out regression analysis with the walnut weight recorded under dynamic conditions as a dependent variable to obtain a linear correlation between the two, and use this as the walnut weight The prediction model calculates the predicted value, and the relative error of the walnut weight is obtained by dividing the absolute value of the difference between the predicted value of the walnut weight and the test value by the test value. 6) The golden section optimization algorithm finds out the best discriminant threshold: take the weight relative error of the tested sample as the independent variable, and the discriminant accuracy rate when each relative error is the threshold is the dependent variable, and a regression analysis is carried out to obtain both There is a Pulse function relationship between them. This function is a unimodal function. The maximum value in the interval of the independent variable is calculated by using the golden section method. At this time, the value of the corresponding independent variable is the optimal threshold. 7) Discrimination method of shriveled walnut using the best threshold value: the walnut image under online conditions is digitally processed and calculated to obtain the projected area of the walnut, the weight of the walnut is predicted by the projected area, and the relative error of the walnut weight is obtained by calculating with the walnut test weight. Comparing the obtained optimal threshold value, if the relative error is greater than the threshold value, it is judged as shriveled walnut, otherwise it is normal walnut.