This study proposes an improved YOLOv5s model, SBCS-YOLOv5s, for accurate and fast detection of tomatoes in natural environments. The model integrates several modules to enhance performance: SE attention, BiFPN, CARAFE, and Soft-NMS. The SE attention module improves feature extraction by focusing on important information, while BiFPN enhances feature fusion across different scales. CARAFE improves feature maps with richer semantic information, and Soft-NMS enhances detection accuracy for overlapping and occluded fruits. The model achieves a mean average precision (mAP) of 87.7% with an IoU of 0.5 to 0.95, which is 3.5% higher than the original YOLOv5s model. It also has a detection speed of 2.6 ms per image, outperforming other state-of-the-art detection algorithms. The model was tested under various conditions, including different lighting, occlusion, and overlap scenarios, demonstrating robust performance. The results show that SBCS-YOLOv5s is effective in accurately detecting tomatoes in natural environments. The study concludes that the model offers improved accuracy, efficiency, and robustness for tomato detection. Future work includes incorporating contextual information and tomato maturity data to further enhance detection accuracy.This study proposes an improved YOLOv5s model, SBCS-YOLOv5s, for accurate and fast detection of tomatoes in natural environments. The model integrates several modules to enhance performance: SE attention, BiFPN, CARAFE, and Soft-NMS. The SE attention module improves feature extraction by focusing on important information, while BiFPN enhances feature fusion across different scales. CARAFE improves feature maps with richer semantic information, and Soft-NMS enhances detection accuracy for overlapping and occluded fruits. The model achieves a mean average precision (mAP) of 87.7% with an IoU of 0.5 to 0.95, which is 3.5% higher than the original YOLOv5s model. It also has a detection speed of 2.6 ms per image, outperforming other state-of-the-art detection algorithms. The model was tested under various conditions, including different lighting, occlusion, and overlap scenarios, demonstrating robust performance. The results show that SBCS-YOLOv5s is effective in accurately detecting tomatoes in natural environments. The study concludes that the model offers improved accuracy, efficiency, and robustness for tomato detection. Future work includes incorporating contextual information and tomato maturity data to further enhance detection accuracy.