A suite of novel image processing is introduced such as the techniques of object detection, semantic segmentation, and object tracking. Particularly, the use of a depth sensor allows for the enhancement of image information by providing a corresponding depth matrix for each RGB image. The agricultural product experimented with herein is the carrot, which has a range of sizes/ shapes and is widely produced in many countries. Through extensive experimentation and analysis, the proposed framework demonstrated high accuracy in estimating the length, width, and volume of carrots, with average errors of 1.85%, 2.51%, and 5.35% when tested on a sample of 20 carrots and a total of 3120 images.
[2022] An automated cucumber inspection system based on neural network
The developed system incorporates both the software and hardware components, in which the geometric properties of a moving cucumber on a conveyor belt can be computed. Concretely, an industrial camera is employed to capture the image of a cucumber. Then, three individual detection systems that perform the cucumber identification, geometry properties approximation, and defect detection, are designed. Finally, if the cucumber is found defective, the PLC motor control will be activated to separate the cucumber into an alternative container. As a result, the proposed algorithms yield promising performances when experimenting on a selfcollected data set, namely “Cuc-70” that consists of a total of 4620 images. The cucumber identification generates an average WIoU of 93%, volume approximation accuracy of 98%, and defect detection WIoU of 92%.
In this study, a two-level Gaussian pyramid is applied to decompose raw data into different resolution levels simultaneously filtering the noises to acquire compact and representative features. Subsequently, a multi-receptive field fusion-based network (MRFFN) is developed to learn the hierarchical features and synthesize the respective prediction scores to form the final recognition result. As a result, the proposed method is capable of exhibiting an outstanding performance of 99.75% when trained using a lightweight dataset. In addition, the experiments conducted using the disturbance defect dataset showed the robustness of the proposed MRFFN against common noises and motion blur.