EfficientNet-based nut surface defect classification method

Abstract

The invention discloses a EfficientNet-based nut surface defect classification method, and the method comprises the steps: 1, collecting a training sample, and making a data set; training an EfficientNet model through the data set, and carrying out nut surface classification after model training convergence; self-adjusting the model scale by using the EfficientNet, so that the problems that time is wasted by manual adjustment and an optimal solution is difficult to obtain are avoided; 2, collecting a target image, and cutting the target image; step 3, modifying a loss function into focal loss, adjusting positive and negative sample weights, and improving model precision; avoiding the influence of imbalance of positive and negative samples on the model; step 4, image preprocessing: making a template image to perform rotation correction on the input image, performing subtraction to obtain a difference image, and fusing the three images into a three-channel image as a model input image; step 5, performing model reasoning: performing reasoning on an input image by using an EfficientNet model, and performing prediction on the input image by using the EfficientNet model, so as to enhance the adaptability of the model to a production environment and improve the generalization ability of the model; and step 6, outputting the reasoning image to an upper computer for display.

Description

technical field [0001] The invention relates to the technical field of nut surface defect classification, in particular to a nut surface defect classification method based on EfficientNet. Background technique [0002] With the development of science and technology, the automation level of industrial production is also increasing. In nut production, while improving production efficiency, how to improve detection efficiency has become an urgent problem to be solved; currently, most factories use manual inspection for quality inspection, which is affected by factors such as staff experience and working status Influenced by the lack of objectivity and heavy workload, the detection efficiency is low; traditional visual inspection methods used in nut defect detection algorithms usually have low accuracy, poor generalization ability and difficult to adapt to the factory environment. Therefore, the nut defect classification technology based on EfficientNet has important practical ...

AI Technical Summary

Problems solved by technology

[0003] At present, most factories use manual inspection to detect surface defects of nuts, which is inefficient

However, the nut surface defect detection algorithm based on traditional vision is susceptible to interference from external environment and other factors, with low accuracy and poor generalization ability.

Images