Garbage multi-target detection method based on improved YOLOv3

Abstract

The invention discloses a garbage multi-target detection method based on improved YOLOv3. In the convolutional neural network, the neuron receptive field of the model is expanded by using 7 * 7 convolution with the step length of 2, so that extraction of shallow information is facilitated; a deeper backbone network framework is constructed by using dense convolution blocks, so that a plurality oftargets difficult to distinguish can be identified, and meanwhile, the phenomenon of overfitting easily occurring when the model is trained to a deep network can be reduced due to the parameter transfer characteristic of dense convolution; on the basis of the whole framework, in cooperation with adjustment of training parameters, the whole model can be used for optimizing a multi-target detectiontask in a complex scene, the detection precision is better than that of a traditional model, and a thought and a method for solving problems are provided for intelligent treatment of household garbageand multi-target detection.

Description

technical field [0001] The invention relates to a garbage multi-target detection method based on improved YOLOv3, which belongs to the fields of artificial intelligence garbage classification and computer vision. Background technique [0002] At present, the common target detection algorithms in the field of computer vision mainly include two categories: one is represented by R-CNN, SSP-net and R-FCN for target classification by generating region candidate frames; the other is based on YOLO and SSD are representative algorithms based on end-to-end detection of the entire picture. In contrast, although the end-to-end algorithm has an advantage in speed, there is still room for improvement in detection accuracy. [0003] For the detection of garbage targets in complex backgrounds, some scholars use the method of segmenting the image background first and then performing target detection to alleviate the interference of complex backgrounds, but it is difficult to meet the nee...

AI Technical Summary

Problems solved by technology

[0003] For the detection of garbage targets in complex backgrounds, some scholars use the method of segmenting the image background first and then performing target detection to alleviate the interference of complex backgrounds, but it is difficult to meet the needs of multiple different backgrounds; use random forest classifiers to extract image features at the pixel level and Fusion can effectively improve the effect of garbage detection, but it is limited to binary classification problems; it is common to optimize the YOLOv3 pre-selection box by clustering algorithm to improve the detection accuracy, but due to the unstable characteristics of domestic garbage, outliers are prone to interfere with the clustering effect, making Accuracy improvement is not obvious

Images