Design and experiment of fresh corn quality detection classifier based on machine vision

In the deep processing of fresh corn, the detection and classification of fresh corn quality are an important but tedious process.The traditional treatment needs a lot of experienced workers to complete this operation, while the results of product quality detection and classification are affected by the subjective experience factors.In order to ensure product quality and increase productivity, an automatic detection and classification algorithm for corn product quality and the equipment are designed in this paper.This automatic device consists of the vision acquisition module, detection and classification control module and execution control module.The vision acquisition module acquires the images of products through the cameras which are installed on the device.In the equipment design process, 2 work stations are designed to capture the images of fresh corn in different view, and the position of fresh corn product is rolled over by a designed mechanical device which is driven by a step motor.In order to provide high quality images for the detection and classification, a light emitting diode (LED) light source is installed near the camera and lighting the measured product during the process of image acquisition; the detection and classification control module is the control core, and it accomplishes the image filtering, texture feature extraction and product classification.In the end of detection and classification operation, this module will export the control instruction to the execution control module.The execution module consists of motion control card, servo controller and servo motor.Using these components, this module moves for special degrees according to the control instruction, and sends the measured products to the designated storage location.In the detection and classification algorithm design process, the computer vision technology is used to detect the fresh corn images and extract texture feature of image.At first, we capture the fresh corn images from different angles of view, and then the texture features of fresh corn images are calculated through the wavelet analysis.In this algorithm, the high-frequency components of texture feature in the horizontal, vertical and diagonal direction are calculated respectively; based on the analysis of texture feature, the separation degree of texture feature is measured by the maximum visual entropy function.The difference between different categories is obtained by comparing with the texture feature of corn products.At the same time, the weight criterion is coordinated with the vision detection algorithm to complete the product detection and classification.According to the texture feature and entropy criterion, the fresh corn products are divided into the following 5 categories, which are peel not clear, pest pollution, no grain damage, minor grain damage and serious grain damage.The equipment prototype has been tested in laboratory, and the experiment results show that the different species, size and broken level of fresh corn can be detected and classified using this device, and the pest pollution product can be eliminated effectively.The average speed of quality detection and classification reaches 1 500 kg per hour, and the effective composite classification rate reaches 99% for the fresh corn with different weight and size; for different damaged corn products, the effective classification rate reaches 92%, which is lower than the composite classification rate.The analysis finds that it is caused by the image acquisition environment and graininess of entropy criterion in the image processing algorithm.In the process of image acquisition, the irradiating degree and illumination of LED light source affect the quality of images.At the same time, the thicker the entropy graininess, the coarser the product detection and classification; the finer the entropy graininess is, the more time the calculation spends.In the future, we will carry out the research on the image acquisition environment construction and the entropy criterion parameter optimization.