System and method for detecting apple surface defect

Abstract

The invention discloses a system and a method for detecting apple surface defects. The system comprises a transmission device (1), and a light box (2) right above the transmission device; a visible /near-infrared composite LED light source device (3) is arranged at the lower part in the light box (2), and a lattice near-infrared light projector (7) is arranged at the upper central part in the light box (2); one side of the lattice near-infrared structure light projection (7) is provided with a black and white industrial camera (4) and a color industrial camera (6); the black and white industrial camera (4) is connected to a near-infrared image acquisition card (8) of a computer (10) through cable; and the color industrial camera (6) is connected to a color image acquisition card of the computer (10) by cable.

Description

Technical field [0001] The invention relates to the technical field of vegetable and fruit detection, in particular to an apple surface defect detection system and method. Background technique [0002] At present, the post-production processing level of apples in my country is very low, and the post-production classification processing technology of apples is seriously backward, which restricts the development of my country's apple industry. The apple external quality sorting machine based on machine vision can comprehensively detect the main indicators of apple size, color, fruit shape and surface defects. It has the advantages of automatic, efficient, accurate, comprehensive, objective and non-destructive inspection, and it is graded after apples. It has been widely used in processing. [0003] The automatic detection methods for apple size, shape, and color indicators are relatively mature, and the rapid identification of defects has always been an obstacle to real-time grading...

AI Technical Summary

Problems solved by technology

[0004] Feature detection method: extract the features of defect area and fruit stem / calyx area, and use artificial neural network, support vector machine and other classifiers to identify. The disadvantage of this type of method is that the recognition accuracy is low

[0005] Structured light detection method: use the deformation of the strip structured light on the surface of the apple to obtain the three-dimensional information of the apple, so as to determine the sunken area as the fruit stem / calyx area. The main disadvantage of this method is that when the fruit stem / calyx area is different from the structure When the directions of the light projectors are consistent, the bar-shaped structured light has only a very small deformation, which makes it impossible to detect the sunken area on the apple surface

[0006] Hyperspectral detection method: Use hyperspectral cameras to detect surface defects on apples. Although the recognition accuracy is high, hyperspectral cameras are extremely expensive and difficult to use in actual production.

[0007] Shadow reconstruction detection method: use the shadow on the apple surface to reconstruct the apple in 3D, and identify the sunken area on the apple surface according to the reconstruction results. This method is difficult to ensure that the apple can provide enough shadow information for 3D reconstruction when it is in any posture, and when Defective areas are misidentified as fruiting stem / calyx areas when their color characteristics are similar to shading

[0008] As an active stereo vision method, the structured light method is widely used to obtain 3D information of objects, but the visible structured light is easily interfered by the color information of the apple surface, which affects the accuracy of obtaining 3D information.

In 1996, some scholars used near-infrared strip-shaped structured light to assist in obtaining three-dimensional information on the surface of apples. However, when the direction of the fruit stem / calyx and the structured light projector are in the same direction, the deformation of the near-infrared strip-shaped structured light is extremely small, leading to this method. The detection accuracy rate is very low and cannot meet the requirements of actual production

Images