Cotton Science

Figure/Table detail

Research on posture and grade recognition of cotton and its localization using improved YOLOX

Xie Jia, Chen Xuefei, Li Yongguo, Jin Changbing, Liang Jintao, Sun Shuaihao

Cotton Science, 2024, 36(4): 328-339. DOI: 10.11963/cs20240016

类别 Categories	训练集 Training set	测试集 Test set	总数 Total
正面普通棉花 Positive normal cotton	564	62	626
正面低品级棉花 Positive low-grade cotton	498	55	553
侧面棉花 Side cotton	428	47	475
其他白色花朵 Other white flower	90	10	100

Table 2 Class distribution of training set and test set images

Other figure/table from this article

Table 1 Image source for the dataset
Fig. 1 Equipment for field experiment
Fig. 2 Technology roadmap
Fig. 3 Recognition module network structure in this study Focus is the slicing operation module for the input image; C3 is the concentrated comprehensive convolution block; CBS is the convolution unit; Concat indicates feature stitching; Upsample is feature upsampling; Detect is the detection module; Dark is DarkNet, and Dark+CA is the CA module added after the Dark module.
Fig. 4 The structure of CA module C denotes the depth of the feature channel; H and W denote the height and width of the feature; r is the indentation ratio; X Avg Pool denotes horizontal global pooling; Y Avg Pool denotes vertical global pooling; Conv2d denotes convolutional 2D; BN denotes batch normalization; Sigmoid is an activation function
Fig. 5 Prediction box and ground truth box of the model $b_{c_{x}}^{g t}$, $b_{c_{y}}^{g t}$ are the coordinates of the center point B^gt of the real box (ground truth box), and $b_{c_{x}}$, $b_{c_{y}}$ are the coordinates of the center point B of the prediction box; h and w are the length and width of the prediction box, and h^gt, w^gt are the length and width of the real box; σ is the straight line distance between the point B^gt and the point B; c_h and c_w are the distance between the center point of the two boxes in the vertical direction and the distance in the horizontal direction; c_h and c_w are the height and width of the smallest outer rectangle of the real box and the prediction box; α is the angle between the center point of the real box and the center point of the prediction box in the horizontal direction.
Fig. 6 The central point of the prediction box Point P is the central point of the prediction box. The red box is the prediction box. Cotton_Top is the predicted category, where it represents positive normal cotton. 1.0 is the confidence level.
Fig. 7 Schematic diagram of the disparity calculation The figure uses two directional axes, Z and X; O_left and O_right are the left and right camera optical centers, respectively; P is the positioning point on the object to be measured; P_left and P_right are the imaging points of the positioning point P on the left and right camera optical sensors, respectively; x_left and x_right are the distances of P_left and P_right from the optical axis of the left and right cameras, respectively; f is the camera focal length; b is the distance between the centers of the two cameras; x is the coordinate of point P on the X-axis; Baseline is the line connecting the optical centers of the left and right cameras.
Fig. 8 Training loss function for different models
Table 3 Results of ablation study
Table 4 The training results of different models
Fig. 9 Identification test results for different cotton types (A) and cover conditions (B) Cotton_Side represents the side cotton identified based on the YOLOX-Cotton model; Cotton_Top represents the positive normal cotton; CottonLow_Top represents the positive low-grade cotton.
Fig. 10 Field cotton category recognition results based on YOLOX-Cotton Cotton_Side represents the side cotton identified based on YOLOX-Cotton model; Cotton_Top represents the positive normal cotton; CottonLow_Top represents the positive side of low-grade cotton. The number at the end of the cotton grade represents confidence.
Table 5 Field verification results of cotton location depth of YOLOX-Cotton model

模态框（Modal）标题

Figure/Table detail