Identification and Detection of Picking Targets of Orah Mandarin Orange in Natural Environment Based on SSD Model

Abstract

With the development of target recognition technology, the technology has been increasingly applied in agriculture. Focusing on Orah mandarin orange, this study makes a dataset of Orah mandarin orange in the natural environment and optimizes the SSD deep learning target detection model. Considering the growth characteristics of Orah mandarin orange in the natural environment, Mobile-Net_v1 is used as the main network for feature extraction of the SSD model in the feature extraction module, replacing the VGG16 network in the classic SSD detection model. This method improves the model's effect on small target detection. Based on the Keras convolutional neural network framework, this model compares the effects of multiple different model detection experiments on the Orah mandarin orange image in the natural environment. The results show that the improved SSD detection model has a significant improvement in detection accuracy, detection speed, and robustness, which provides a new solution for automatic picking of Orah mandarin orange in the future.