Enhanced mango fruit detection based on edge artificial intelligence

Abstract Accurate detection of mango fruits and their corresponding stems is a key supporting technology for automating mango harvesting. However, factors such as occlusion by leaves and branches, overlapping of clustered fruits, and complex and variable lighting conditions in natural scenes significantly increase the difficulty of target detection. To mitigate the impact of these diverse factors in natural environments, this study proposes an improved target detection model based on YOLO11n for detecting mango fruits and their stems. Firstly, to expand the receptive field of the convolution and effectively capture low-frequency information in the image, we integrate wavelet convolution with the C3k2 module to construct the WT_C3k2 module. Secondly, we propose a Fuzzy Channel Spatial Attention Mechanism (FCSA) module, which aims to enhance the model’s performance and robustness in handling ambiguous, uncertain, or complex data. Finally, we replace the original Feature Pyramid Network (FPN) with a Bidirectional Feature Pyramid Network (BiFPN), enabling more comprehensive flow of feature information. This facilitates better capture of features across different scales and complex backgrounds while effectively reducing redundant computations. The experimental results demonstrate that the improved model achieves a detection accuracy of 98.5% for mango fruits and 94.2% for stems, outperforming the original model. Additionally, we deployed the proposed model on an edge device, achieving an inference speed of 41.6 frames per second. The detection model presented in this study can be effectively applied to intelligent devices such as harvesting robots, providing technical support for automated mango harvesting.