果实目标深度学习识别技术研究进展

机器视觉技术是果实目标识别与定位研究的关键。传统的目标识别算法准确率较低、检测速度较慢，难以满足实际生产的需求。近年来，深度学习方法在果实目标识别与定位任务中表现出了优良的性能。本文从数据集制备与果实目标识别模型两方面进行综述，总结了数据集制备相关的有监督、半监督和无监督3种方法的特点，按照深度学习算法的发展历程，归纳了基于深度学习的果实目标检测和分割技术的常用方法及其实际应用，轻量化模型的研究进展及其应用情况，基于深度学习的果实目标识别技术面临的问题和挑战。最后指出基于深度学习的果实目标识别方法未来发展趋势为：通过弱监督学习来降低模型对数据标签的依赖性，提高轻量化模型的检测速度以实现果实目标的实时准确检测。

Review on Deep Learning Technology for Fruit Target Recognition

Machine vision technology is the key of fruit target recognition and positioning. Traditional target recognition algorithm has low accuracy and slow detection speed, which is difficult to meet the needs of actual production. In recent years, deep learning methods have shown excellent performance in fruit target recognition and localization tasks. The fruit target recognition algorithm based on deep learning has the advantages of high detection progress and fast detection speed, so it is widely used in the fruit target recognition task under different scenes and has achieved many good achievements. The data set preparation and fruit target recognition models were reviewed. Firstly, the characteristics of supervised, semi-supervised and unsupervised methods related to dataset preparation were summarized. Secondly, according to the development process of deep learning algorithm, the common methods and practical applications of deep learning-based fruit target detection and segmentation technology were summarized, the previous research on the detection and segmentation of fruit objects such as apple, citrus and tomato under different natural scenes was summarized, and the research progress and application of lightweight model were summarized. Thirdly, the problems and challenges of deep learning-based fruit target recognition technology were summarized. In the end, the future development trend of deep learning-based fruit target recognition methods was pointed out, that was, weakly supervised learning would be used to reduce the dependence of models on data labels, and the detection speed of lightweight models would be improved to achieve real-time and accurate detection of fruit targets.