Research on adaptive object detection via improved HSA‐YOLOv5 for raspberry maturity detection

Abstract In the field of machine vision, target detection models have experienced rapid development and have been practically applied in various domains. In agriculture, target detection models are commonly used to identify various types of fruits. However, when it comes to recognizing berries, such as raspberries, the fruits nearing ripeness exhibit highly similar colours, posing a challenge for existing target detection models to accurately identify raspberries in this stage. Addressing this issue, a raspberry detection method called HSA‐YOLOv5 (HSV self‐adaption YOLOv5) is proposed. This method detects immature, nearly ripe, and ripe raspberries. The approach involves transforming the RGB colour space of the original dataset images into an improved HSV colour space. By adjusting corresponding parameters and enhancing the contrast of similar colours while retaining the maximum features of the original image, the method strengthens data features. Adaptive selection of HSV parameters is performed based on data captured under different weather conditions, applying homogeneous preprocessing to the dataset. The improved model is compared with the original YOLOv5 model using a self‐constructed dataset. Experimental results demonstrate that the improved model achieves a mean average precision (mAP) of 0.97, a 6.42 percentage point increase compared to the baseline YOLOv5 model. In terms of immature, nearly ripe, and ripe raspberries, there are improvements of 6, 4, and 7 percentage points, respectively, validating the effectiveness of the proposed model.