The invention provides a method and equipment for recognizing multiple attributes of target objects, a medium and a neural network. According to the method, a plurality of attribute features of a target object can be synchronously recognized by a plurality of full connection layers, and the quantity of the full connection layers can be set according to the quantity of the to-be-recognized attribute features, so that the modification is convenient and the flexibility is high. The layer number of the neural network does not exceed 34, so that the processing time of a single picture can be rapidly shortened while the precision is ensured.

Method and equipment for recognizing multiple attributes of target objects, medium and neural network

The invention discloses a method for realizing convolution neural network based on a memristor, which comprises the following steps: preparing a convolution neural network to be trained composed of convolution + pooling layer and full connection layer, inputting image information of a training set to obtain output value of a full connection layer; according to the deviation between the output value of all connection layers of the convolution neural network to be trained and the standard information, the convolution neural network to be trained being trained by the back propagation function; when the training times are reached, whether the training precision is up to the standard or not being judged If the training precision is not up to the standard, the convolutional neural network beingtrained again until the precision is up to the standard A method for obtain a desired convolution neural network and that like The remarkable effects are: the influence on precision, difficulty ofpreparation and time is reduced in hardware; a more advanced multi-layer convolution neural network is realized The method for realizing convolution neural network based on a memristor is not easy tobe affected by hardware noise, preparation difficulty and time, and after the application of complex can show a stronger adaptive capacity

A method for realizing convolution neural network based on memristor

The invention provides an image processing method and device based on a full convolutional neural network, and the method comprises the steps: carrying out the data refinement of the image data of a to-be-processed image, converting the to-be-processed image subjected to the data refinement into a single-channel gray level image corresponding to three primary colors, carrying out the image convolution, and obtaining a feature map of the to-be-processed image; and performing image deconvolution on the feature map of the to-be-processed image to restore the feature map to a target image which has the same resolution as the to-be-processed image and marks the category of the target object. The data refinement enables the pixel value of each pixel point in the image to be more accurate, and more detail features can be extracted when features are extracted through convolution. The three-channel image is converted into the three single-channel grayscale images corresponding to the three primary colors for convolution, and compared with direct convolution of the three-channel image, the loss of image features can be reduced, so that the final image processing result is more accurate, and the method can be well used for processing the image under the weak illumination condition.

Image processing method and device based on full convolutional neural network

The invention discloses a foreign matter detection method for an overhead transmission line based on image processing. The method comprises the following steps: shooting three resolution images of each overhead transmission line containing foreign matters; Preprocessing each image; Setting a corresponding foreign matter label for each preprocessed image; Constructing a multi-scale convolutional neural network model; Taking the three resolution images corresponding to each foreign matter label as input values in parallel, and inputting the input values into a multi-scale convolutional neural network model; Training the convolutional neural network model to obtain a trained convolutional neural network model; Shooting and preprocessing the overhead transmission line image to be detected in the same way, and inputting three resolution images of the overhead transmission line to be detected into the trained convolutional neural network model in parallel; And outputting the foreign matter type in the overhead transmission line image to be detected.

A foreign matter detection method for an overhead transmission line based on image processing

The invention discloses a crop disease recognition method based on a CNN (Convolutional Neural Network) and a multi-video image, and belongs to the technical field of image processing and machine learning. The method comprises the steps: expanding an original color disease leaf image training set collected through the agricultural IOT (Internet of Things), and obtaining an expanded training set; constructing a multi-channel CNN, training the multi-channel CNN through the expanded training set, and obtaining a crop disease recognition model; inputting a crop disease leaf image of a to-be-recognized crop into a crop disease recognition model for image classification, and obtaining the type of the disease of the to-be-recognized crop. The method can improve the accuracy of the crop disease recognition model, and can solve a problem that the accuracy of a conventional crop disease recognition method is lower in accuracy.

Crop disease recognition method based on CNN (Convolutional Neural Network) and multi-video image

The invention relates to a crop disease recognition algorithm based on a three-channel convolutional neural network. The crop disease recognition algorithm is mainly and technically characterized by comprising the steps image preprocessing is conducted; RGB colored diseased leaf images are decomposed into an R channel, a G channel and a B channel to serve as inputs of the convolutional neural network; a three-channel convolutional neural network model is constructed; weights and offset parameters are initialized; and the trained convolutional neural network is retrained through a training database, therefore, the trained model parameters can be fit with a current database, and more abstract features of the images are extracted for image recognition. The crop disease recognition algorithm is reasonable in design, experiment verification is conducted on an apple diseased leaf image database and a cucumber diseased leaf image database, cucumber diseased leaf image features can be accurately described, therefore, the obtained feature values of the target pixel can be more representative, and the recognition rate is increased; and description of the obtained feature values on feature information of cucumber diseased leaf images is comprehensive, and therefore cucumber diseased leaf image recognition can be effectively conducted.

Crop disease recognition algorithm based on three-channel convolutional neural network

Efficient and accurate recognition of plant diseases based on leaf images is a hot research topic. The plant diseased leaf images are complex and diverse. It is generally difficult to extract reliable features. In this paper, a new plant disease recognition method is proposed, based on a Modified Residual Neural Network (MRNN) and transfer learning. Compared with the classical residual neural network ResNet-50, the residual block structure in MRNN is modified. The experiment results on the AI Challenger dataset show MRNN can achieve 91.4% recognition accuracy which is higher than other classic CNN models. Combined with the Kaggle Cassava dataset, the MRNN is trained with transfer learning, which improves the accuracy, robustness and generalization ability. The experiments results show that the proposed method not only has an advantage in accuracy, but also has a significant improvement in training speed, which validates the efficiency and effectiveness of the proposed approach.

Jucheng Yang

Papers

Crop disease recognition algorithm based on three-channel convolutional neural network

Efficient Plant Diseases Recognition based on Modified Residual Neural Network and Transfer Learning