The detection of banana fruits is an important part of intelligent management in the banana plantation. To detect the banana fruit quickly and accurately in the complex orchard environment, this paper proposes a method based on the latest deep learning algorithm to detect the banana fruit. Using a monocular camera, we applied the YOLOv4 neural network algorithm to extract the deep features of banana fruits, realizing accurate detection of different banana sizes. The detection algorithm achieved a 99.29% detection rate, the average execution time was 0.171s, the shortest execution time was 0.135s, and the AP was 0.9995. Moreover, the detection results were discussed with the YOLOv3 algorithm and the machine learning algorithm. Compared with the machine learning algorithm, deep learning algorithm was superior to both detection accuracy and detection time. YOLOv4 had higher detection confidence and higher detection rate than YOLOv3. The results show that the proposed method could realize the fast detection of different varieties and different maturity in banana plantations, under different illumination and occlusion conditions, and provide information for banana picking, maturity and yield estimation. INDEX TERMS Banana detection, orchard environment, deep learning, green fruit, YOLOv4. L.Fu et al.: Fast and Accurate Detection of Banana Fruits in Complex Background Orchards L.Fu et al.: Fast and Accurate Detection of Banana Fruits in Complex Background Orchards
The real-time detection of banana bunches and stalks in banana orchards is a key technology in the application of agricultural robots. The complex conditions of the orchard make accurate detection a difficult task, and the light weight of the deep learning network is an application trend. This study proposes and compares two improved YOLOv4 neural network detection models in a banana orchard. One is the YOLO-Banana detection model, which analyzes banana characteristics and network structure to prune the less important network layers; the other is the YOLO-Banana-l4 detection model, which, by adding a YOLO head layer to the pruned network structure, explores the impact of a four-scale prediction structure on the pruning network. The results show that YOLO-Banana and YOLO-Banana-l4 could reduce the network weight and shorten the detection time compared with YOLOv4. Furthermore, YOLO-Banana detection model has the best performance, with good detection accuracy for banana bunches and stalks in the natural environment. The average precision (AP) values of the YOLO-Banana detection model on banana bunches and stalks are 98.4% and 85.98%, and the mean average precision (mAP) of the detection model is 92.19%. The model weight is reduced from 244 to 137 MB, and the detection time is shortened from 44.96 to 35.33 ms. In short, the network is lightweight and has good real-time performance and application prospects in intelligent management and automatic harvesting in the banana orchard.
Litchi downy blight caused by Peronophythora litchii is the most serious disease in litchi production, storage and transportation. Existing disease identification technology has difficulty identifying litchi downy blight sufficiently early, resulting in economic losses. Thus, the use of diffuse reflectance spectroscopy to identify litchi downy blight at different stages of disease, particularly to achieve the early identification of downy blight, is very important. The diffuse reflectance spectral data of litchi fruits inoculated with P. litchii were collected in the wavelength range of 350–1350 nm. According to the duration of inoculation and expert evaluation, they were divided into four categories: healthy, latent, mild and severe. First, the SG smoothing method and derivation method were used to denoise the spectral curves. Then, the wavelength screening methods competitive adaptive reweighted sampling (CARS) and successive projections algorithm (SPA) were compared to verify that the SPA method was more effective. Eleven characteristic wavelengths were selected, accounting for only 1.1% of the original data. Finally, the characteristic wavelengths were tested by six different classification models, and their accuracy was calculated. Among them, the ANN model performed best, with an accuracy of 90.7%. The results showed that diffuse reflectance spectroscopic technology has potential for identifying litchi downy blight at different stages, providing technical support for the subsequent development of related automatic detection devices.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.