A Symbiosis with Panicle-SEG Based CNN for Count the Number of Wheat Ears

The number of wheat ears per unit area counting helps to measure the wheat grain quality. The manual counting process is a high laborious and time-consuming process. For better counting of wheat ears, a Panicle segmentation-based CNN approach has been used to estimate the number of wheat ears in the wheat plant. A total of 236 wheat ears have been taken from the wheat ears detection dataset [1]. The manually labeled and superpixels data have been integrated to form segmented data. The segmented data have been used for the classification of wheat ear classes. As the 1056 segmented panicles have been extracted. The segmented panicles were constructed into three classes: one ear, two ears, and three ears and these panicles have been used for training and validation purposes in CNN. The panicle segmentation-based CNN approach achieves a high F1 score (98.3%) and Q
seg
(0.82) in one ears classification. Classification of segmented images, unlike target recognition, not only reduces the labour of manual annotation but also greatly improves the efficiency and accuracy of wheat ear counting, thereby meeting the requirements of wheat yield estimation in the field environment.