Fruit classification plays a pivotal role in the fruit industry. Traditional fruit classification methods rely heavily on manual inspection, which significantly hampers the accuracy and efficiency of harvesting. The advent of deep learning algorithms has empowered robots to classify fruits more swiftly, thereby enhancing picking efficiency. In this paper, we propose a deep optimization target detection algorithm based on the YOLOv8s framework to improve the accuracy and efficiency of fruit classification. This algorithm leverages the YOLOv8s model and introduces optimizations to enhance fruit classification effectiveness. Firstly, we focused on making crucial enhancements to the head and neck network's structure. Specifically, we replaced the original C2f module with the Enhanced Local Aggregation Network (ELAN) module, which retains more comprehensive feature information and learns directly from the original feature map. This modification effectively minimizes information loss, bolsters feature representation capabilities, and subsequently elevates detection accuracy. Secondly, we integrated receptive field attention into both the convolutional layer and feature extraction layers of the backbone and neck networks. By augmenting the network's capacity to capture local region and global context information, this design significantly improves the model's efficiency in learning detailed features, thereby accelerating target detection speed and refining detection accuracy. Furthermore, in fruit classification detection tasks, incorporating multiple attention mechanisms before the backbone network and the large object detection head enables dynamic adjustment of feature weights. This approach strengthens key target features, suppresses background noise, and markedly reduces the false detection rate. This effect is particularly pronounced in scenarios with numerous fruit picking targets and complex backgrounds. The proposed model exhibits excellent detection accuracy and is suitable for practical fruit detection applications
