Image Cropping Method

To solve the loss of small target features in the YOLO algorithm during input image size adjustments, an image cropping method, Size-preserving Image Subdivision and Concatenation (SISC), is proposed. This method uses predefined dimensions to determine the boundaries of dataset images. Sub-images will be generated by conducting horizontal and vertical cropping across the entire image. Then, it also adjusts the original annotation boxes to accommodate 16 potential adjustment scenarios, generating sub-annotation box files corresponding to the sub-images. Additionally, the Grayscale Omission of Peripheral features (GOP) method is proposed to enhance target features. This method traverses through the pixels of the image based on the boundaries of the original annotation boxes, dividing the image into important and non-important regions, and calculates the (R,G,B) mean values of the non-important regions as grayscale values for replacement. The efficacy of the proposed method is tested using the CropAndWeed dataset with YOLO v5 and YOLO v8 algorithms. The results show that the combined application of SISC and GOP significantly improves YOLO’s precision (P), recall (R), and mean average precision (map (0.50:0.95)) by at least 8.7%, 11.1%, and 14.5%, respectively.