基于改进CBAM注意力机制的MobileNetV2玉米种子品种识别研究

Mobile NetV2 Maize Seed Variety Recognition Based on Improved Attention Mechanism CBAM

玉米是我国主要粮食作物,有较高的营养价值和经济价值。不同的地域环境适宜种植的玉米品种不同,但由于玉米种子在外形方面存在的差异较小,所以仅凭肉眼很难对其进行快速准确的识别。为实现玉米种子品种的准确识别,研究采集了9种玉米种子图像共2792张建立数据集,并按照7∶2∶1的比例随机划分训练集、验证集和测试集。将注意力机制CBAM引入轻量化模型MobileNetV2,对CBAM的串行方式进行改进,构建一个新型注意力模块E_CBAM,并通过对比不同的压缩比,选出效果最佳的压缩比为4,提出了E_CBAM_MobileNetV2模型。实验表明E_CBAM_MobileNetV2的准确率为98.18%,相较于MobileNetV2提高了5.45%。

Maize is a main food crop in China,with high nutritional value and economic value.Different geographical environments are applicable for different varieties of maize for planting,however,due to the slight differences in appearance of different varieties,it is difficult to quickly and accurately identify different varieties with the naked eye.In order to realize the accurate identification of maize seed varieties,total 2792 images of 9 kinds of maize seeds were collected in this study to establish a data set,and training set,verification set and test set were randomly divided according to a ratio of 7∶2∶1.In this study,the attention mechanism CBAM was introduced into the lightweight model MobileNetV2,the serial mode of CBAM was improved,a new attention module E_CBAM was built,and different reduction ratios were compared,the best reduction ratio of 4 was selected,and E_CBAM_MobileNetV2 model was proposed.Experiments indicated,that the accuracy of E_CBAM_MobileNetV2 was 98.18%,5.45%higher than that of MobileNetV2.