Dynamic crop morphology capture aids real-time precision in field spraying decisions. Ultrasonic radar and costly LiDAR can't maximize their high precision. They may slow down decision speed because the current spraying amount control needs to include more information about branches and leaves, which highly relies on the statistical features of regional morphology. Since most agricultural tools in China are medium in size and relatively inexpensive, it makes sense to create a unique identification system that combines deep learning with cheap monocular vision. A crop's growth cycle was a helpful gauge for direct spraying because of the knowledge gained. For this reason, let's examine cotton as an example: This study aims to develop a single-camera acquisition system and propose an algorithm based on the Convolutional Neural Network (CNN) to identify different growth stages of cotton plants. An optimal CNN model structure is chosen among nine possible method structures using an optimization procedure based on the matrix of confusion and recognition efficiency, and its dependability is established by repeatedly switching the training and test sets by the k-fold test concept. This CNN model has an accuracy of 94.27%, an F1-score of 96.39%, a recall of 95.76%, a precision of 95.31%, and a recognition speed of 72.46 ms per picture. Also, the convolution neural network, the model suggested in this study, performs better than those of VGG16 and Alex Net. Finally, spraying deposition tests were conducted over three distinct cotton growth stages to validate the accuracy of the developed identification method and the viability of the sprayed decision-making process based on CNN. The findings of the studies confirm that the ideal spray parameters were applied during distinct growth times correspondingly, leading to an increase of up to 63.24% in the optimum index over the operations that did not discriminate growth periods, yielding an overall increase of 43.29%.