Method for controlling a soil working means based on image processing and related system

[0064]With reference to FIGS. 3A, 3B, 3C, in the invention, such a synthetic soil descriptor is a soil classification. In this case, the convolutional neural network 300 or first convolutional network 300 comprises at least the following layers:

[0065]an input layer 301 configured to receive the entire digital image or at least one down-sample of such a digital image acquired with the cameras 203;

[0066]at least one convolutional layer conv 1;

[0067]at least one fully connected layer 303a;

[0068]an output layer 304 with at least one neuron configured to make the distinction between at least two soil classes available.

[0069]In greater detail, the network 300 which operates on the basis of soil classification comprises a convolution block 302 consisting of three convolutional layers conv 1, conv 2, conv 3 in cascade and of three Max Pooling layers MP1, MP2, MP3, of the type known to a skilled in the art.

[0070]Each Max Pooling layer MP1, MP2, MP3 is directly connected to the output of a r...

[0078]With reference to FIG. 4A, in the control method 100, the synthetic soil descriptor is a soil semantic segmentation. In this case, the convolutional neural network 400 or second convolutional neural network 400 comprises at least the following layers:

[0079]an input layer 401 configured to receive the entire digital image or at least one down-sample of the digital image acquired with the cameras 203;

[0080]at least one convolutional layer conv 1;

[0081]at least one deconvolutional layer deconv 1;

[0082]an output layer 404 configured to make a soil image semantically segmented in at least two classes available.

[0083]In greater detail, the network 400 which operates on the basis of soil semantic segmentation comprises a respective convolution block 402 consisting of a plurality of convolutional layers, for example n layers, conv 1, . . . , conv n in cascade, spaced out by n−1 Max Pooling layers MP1, . . . MPn−1.

[0084]Furthermore, the neural network 400 comprises a respective deconvo...

[0090]With reference to FIG. 5A, in the invention, the synthetic soil descriptor is a specific action to be taken by means of working means based on the characteristics of the soil framed.

[0091]In this case, the convolutional neural network 500 or third convolutional neural network 500 is substantially analogous to the convolutional network 300 and comprises at least the following layers:

[0092]an input layer 501 configured to receive the entire digital image or at least one down-sample of the digital image acquired;

[0093]at least one convolutional layer conv 1;

[0094]at least one fully connected layer 503a;

[0095]an output layer 504 with at least one neuron configured to make the distinction between at least two specific actions to be taken on the soil available.

[0096]In greater detail, the network 500 comprises a respective convolution block 502 consisting of three convolutional layers conv 1, conv 2, conv 3 in cascade and of three Max Pooling layers MP1, MP2, MP3. Each Max Pooling ...