Lightweight convolutional neural network models for semanticsegmentation of in-field cotton bolls

作     者：Naseeb Singh V.K.Tewari P.K.Biswas L.K.Dhruw 

作者机构：Department of Agricultural and Food EngineeringIIT KharagpurKharagpur 721302India Department of Electronics and Electrical Communication EngineeringIIT KharagpurKharagpur 721302India 

出 版 物：《Artificial Intelligence in Agriculture》 (农业人工智能（英文）)

年 卷 期：2023年第8卷第2期

页      面：1-19页

核心收录：

学科分类：0710[理学-生物学] 0202[经济学-应用经济学] 02[经济学] 020205[经济学-产业经济学] 0802[工学-机械工程] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 0702[理学-物理学] 

基　　金：Indian Institute of Technology Kharagpur  IIT KGP 

主　　题：Cotton Semantic segmentation Convolutional neural network Robotic harvesting Image segmentation Deep learning 

摘      要：Robotic harvesting of cotton bolls will incorporate the benefits of manual picking as well as mechanical harvesting. For robotic harvesting, in-field cotton segmentation with minimal errors is desirable which is a challengingtask. In the present study, three lightweight fully convolutional neural network models were developed for thesemantic segmentation of in-field cotton bolls. Model 1 does not include any residual or skip connections,while model 2 consists of residual connections to tackle the vanishing gradient problem and skip connectionsfor feature concatenation. Model 3 along with residual and skip connections, consists of filters of multiplesizes. The effects of filter size and the dropout rate were studied. All proposed models segment the cotton bollssuccessfully with the cotton-IoU (intersection-over-union) value of above 88.0%. The highest cotton-IoU of91.03% was achieved by model 2. The proposed models achieved F1-score and pixel accuracy values greaterthan 95.0% and 98.0%, respectively. The developed models were compared with existing state-of-the-art networks namely VGG19, ResNet18, EfficientNet-B1, and InceptionV3. Despite having a limited number of trainableparameters, the proposed models achieved mean-IoU (mean intersection-over-union) of 93.84%, 94.15%, and94.65% against the mean-IoU values of 95.39%, 96.54%, 96.40%, and 96.37% obtained using state-of-the-art networks. The segmentation time for the developed models was reduced up to 52.0% compared to state-of-theart networks. The developed lightweight models segmented the in-field cotton bolls comparatively faster andwith greater accuracy. Hence, developed models can be deployed to cotton harvesting robots for real-time recognition of in-field cotton bolls for harvesting.