A multi-branch convolutional neural network with density map for aphid counting. (January 2022)

• Record Type: Journal Article
• Authors: Li, Rui; Wang, Rujing; Xie, Chengjun; Chen, Hongbo; Long, Qi; Liu, Liu; Zhang, Jie; Chen, Tianjiao; Hu, Haiying; Jiao, Lin; Du, Jianming; Liu, Haiyun
• Is Part Of: Biosystems engineering. Volume 213(2022)
• Page Start: 148
• Page End: 161
• Publication Date: 2022-01
• Subjects: Density map -- Aphid counting -- Convolutional neural network -- Deep learning; Bioengineering -- Periodicals; Agricultural engineering -- Periodicals; Biological systems -- Periodicals; Génie rural -- Périodiques; Systèmes biologiques -- Périodiques; 631
• DOI: 10.1016/j.biosystemseng.2021.11.020
• Languages: English
• ISSNs: 1537-5110

Abstract

Abstract : In agriculture, aphids always cause major damage in wheat, corn and rape, which significantly affect the crop yield. Manual aphid counting approaches are often labour-consuming and time-costing for Integrated Pest Management (IPM). In addition, the results of existing aphid counting methods based on computer vision are not satisfactory due to the complex background and the dense distribution. In order to address these problems, a novel multi-branch convolutional neural network (Mb-CNN) with density map for aphid counting is developed in this paper. In this approach, the aphid images are firstly fed into multi-branch convolutional neural networks, which have three branches for extracting the feature maps of different scales. Then, an aphid density map is generated via Mb-CNN, which contains the distribution information of aphids. Finally, the counting of aphids is estimated by using the density map. Experiment results on our dataset demonstrate that our Mb-CNN achieves the performance of 10.22 Mean Absolute Error (MAE) and 12.24 Mean Squared Error (MSE) in the aphid counting, which outweighs the state-of-the-art approaches. Highlights: Domain specific dataset for aphid counting in the field. This dataset may have high application value in practical aphid monitoring. Propose a multi-branch convolutional neural network with density map for aphid counting.