Research On Bottle Mouth Defect Recognition Algorithm For Empty Beverage Bottle Detection Robot

The empty bottle detection robot is a kind of intelligent detection robot based on visual perception,mainly used in the food and beverage industry.The beverage empty bottles are detected all-around before the beverage is filled,and the defective bottle is removed to ensure the safety of filling and the beverage quality.The empty bottle detection robot is a kind of high-speed and high-precision intelligent manufacturing equipment,and the overall performance of the equipment directly affects the production efficiency of the beverage production line.The design of appropriate visual inspection program and the development of high-speed and high-precision image processing algorithm can be one of the key technologies to improve the overall equipment performance.This paper focuses on the image processing algorithm of the bottle mouth detection unit,aiming to develop a set of bottle mouth recognition algorithm which can achieve high performance index.Before recognition of the bottle mouth defect,the bottle mouth should be located first.The positioning accuracy of the bottle mouth positioning algorithm is not high and the positioning time is long,which has been a difficult problem.This paper studies the algorithms of the bottle mouth location in the existing literature and summarizes the three factors of the deviation of the positioning results.Then a four-circle bottle mouth positioning method is designed with a result of positioning time less than 15 milliseconds and the deviation of the circle center less than 3 pixels.In addition,this paper defines the circle center accuracy as the ratio of the circular projection peak value of the current positioning circle center to the circular projection peak value of the circle center of the best bottle mouth.By moving the circle center in a small range and calibrating the circle center according to the circle center accuracy,the deviation achieves the positioning accuracy of less than 1 pixel while the circle center of the calibration time is long.After positioning the center of the bottle mouth,we will expand the bottle mouth area as the bottle mouth rectangle.However,the highlighted area and the shadow area of the bottle mouth image will interfere with each other,affecting the defect recognition.This paper designs a method of dividing the highlighted area of the bottle mouth.Firstly,by detecting the rising and falling edges of the circular projection curve of the bottle mouth,the highlighted area of the mouth is extracted.Then,subtract the highlighted area from the original image and select pixels in the neighborhood shadow area of the original highlighted area to fill the blank area.Finally,the real reduction image of the non-highlighted bottle mouth is obtained.After extracting the highlighted area of the bottle mouth,the defect detection will be carried out separately for the high-light area and the true reduction of non-highlighted bottle mouth.The radial projection method is used in the highlighted area.Because the defect projection value of the highlighted area is very low,the concave trough detection of the projection curve is carried out.When the length of the concave trough reaches the set threshold value,the trough will be recognized as a defect.The lag threshold segmentation method is employed in the true reduction of non-highlighted bottle mouth.Two thresholds are set,one is used to locate the defect and the other is used to divide the defect area.If the defect area reaches the set size,it will be recognized as a defect.The bottle mouth defect detection algorithm of this paper can achieve recall rate of 99%of bad bottles and the running time is less than 30 milliseconds above the current detection speed 1 of 72000 bottles per hour of international level,which has higher practical value.