Fruit and vegetable image classification system and method

Abstract

The invention discloses a fruit and vegetable image classification system comprising a convolutional neural network used for extracting an input fruit and vegetable image feature map, a low-dimensional SCA attention module used for identifying a low-dimensional key feature map in the low-dimensional feature map of the fruit and vegetable image, a medium-dimensional SCA attention module used for identifying a medium-dimensional key feature map in the medium-dimensional feature map of the fruit and vegetable image, the high-dimensional SCA attention modules used for identifying high-dimensional key feature maps in the high-dimensional feature maps of the fruit and vegetable images, and the pooling layer that is linked with each SCA attention module; the fruit and vegetable image classification system further comprises: a multi-scale feature fusion module which is used for carrying out fusion processing on the low-dimensional key feature map, the medium-dimensional key feature map and the high-dimensional key feature map which are subjected to pooling processing to generate uniform feature representation; and the full connection layer that is used for classifying the fruit and vegetable images according to the unified feature representation.

Description

technical field [0001] The present invention relates to the field of image processing, specifically to the field of fruit and vegetable image classification, and more specifically to the fruit and vegetable recognition technology, that is, a fruit and vegetable image classification system and method. Background technique [0002] Food computing technology has promoted the rapid development of the food industry. As an important branch of food computing, fruit and vegetable recognition has a very wide range of applications in real scenarios. For example, fruit picking robots use fruit and vegetable recognition technology to further improve picking efficiency, and fruit and vegetable recognition technology is used in supermarkets. Carry out intelligent weighing and checkout, and use fruit and vegetable identification technology in household refrigerators for fruit and vegetable quality management. However, there are also three major difficulties in fruit and vegetable image cla...

AI Technical Summary

Problems solved by technology

However, there are also three major difficulties in fruit and vegetable image classification: (1) The same fruit contains a variety of visual information. For example, when describing a pineapple according to a picture, the visual information we can observe includes: yellow fruit Body, green leaves, rough fruit surface, zigzag leaves, etc., all these visual information constitute the main features of the pineapple image, and help us to distinguish pineapple from other fruits, if we only choose to use the A visual information cannot distinguish pineapple as a fruit

(2) There are very large visual differences in the images of fruits and vegetables. For the same fruit, its images may have very large visual differences. Presentation method (some fruits are whole, some are sliced ​​or cut open, and some fruits and vegetables are even packaged), in addition, the number of fruits and vegetables in the image will also cause a huge visual difference in the image

(3) The background of the fruit and vegetable images varies and there is a lot of noise. The background of some fruit and vegetable images often has a lot of information that has nothing to do with the fruit and vegetable itself, such as the branches and leaves they grow or the container in full bloom. Moreover, the fruits and vegetables in the image Can produce huge visual changes due to lighting angles or shadow occlusion, which further increases the difficulty of identification

[0003] From the above description, it can be seen that some existing methods rely on professional equipment for identification, such as near-infrared imagers and tactile sensors, but these equipment are relatively expensive, which is not conducive to the general application of this technology

There are still a large number of methods that directly extract deep visual features for fruit and vegetable image classification through convolutional neural networks (CNN), but they ignore the characteristics of the fruit and vegetable image itself, so it is difficult to achieve the best performance

In addition, most of these methods directly transfer the model of recognizing other objects to the task of fruit and vegetable recognition, without considering the task characteristics of fruit and vegetable recognition

Images